Skip to content

[LoRA] support more comyui loras for Flux 🚨 #10985

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 34 commits into from
Apr 9, 2025

Conversation

sayakpaul
Copy link
Member

@sayakpaul sayakpaul commented Mar 6, 2025

What does this PR do?

Support more ComfyUI Flux LoRAs. Context: #10954 (comment). Depends on huggingface/peft#2419.

This PR has a breaking change explained here:
#10985 (comment)

I have run the LoRA integration tests and they are passing.

@HuggingFaceDocBuilderDev

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@spezialspezial
Copy link
Contributor

This pops up if I do a full text search:

.lora_B.bias -> .diff_b


def convert_lora_bfl_control(sd): #BFL loras for Flux
    sd_out = {}
    for k in sd:
        k_to = "diffusion_model.{}".format(k.replace(".lora_B.bias", ".diff_b").replace("_norm.scale", "_norm.scale.set_weight"))
        sd_out[k_to] = sd[k]

    sd_out["diffusion_model.img_in.reshape_weight"] = torch.tensor([sd["img_in.lora_B.weight"].shape[0], sd["img_in.lora_A.weight"].shape[1]])
    return sd_out

@sayakpaul
Copy link
Member Author

diff_b is lora_B.bias 🧠 thanks!

There are some norm_diff keys, too. Do you have any pointers about them?

@spezialspezial
Copy link
Contributor

You mean text_encoders.clip_l.transformer.text_model.encoder.layers.0.layer_norm1.diff? Seems to be all zeroed out.

When doing a LoRA merge in comfy you subtract the base model from the LoRAed model. Maybe diff is hinting at that and it's some temporaries. Probably doing costly SVD on all that zeros too.

@sayakpaul
Copy link
Member Author

You mean text_encoders.clip_l.transformer.text_model.encoder.layers.0.layer_norm1.diff? Seems to be all zeroed out.

Yes, indeed. I will push the diff_b related fixes soon and update you about that here. the position_embedding stuff also needs a bit of working around but hopefully, we get around to it soon.

I think this particular LoRA in question probably stretches the different LoRA shenanigans the most. But the rest of the ComfyUI LoRAs should hopefully be loadable with this PR as is.

Thanks for helping!

@spezialspezial
Copy link
Contributor

FluxDFaeTasticDetails does load and gives a similar style to what is shown in the LoRA gallery.

Prompt: a majestic magical koi fish swimming gracefully through a serene pond. The koi fish has shimmering, iridescent scales that glow softly, illuminating the water around it, colored ink painting on parchment, black and white ink with splashes of color

sd-0029191-2951100183

@sayakpaul
Copy link
Member Author

Nice, that is promising. Do keep testing other LoRAs (apart from the one we discussed already) and ping me if there's something off.

@BenjaminBossan
Copy link
Member

I am unable to catch why the following would be raised:

If nn.Embedding layers are targeted with LoRA, in PEFT, we use other parameter names, not lora_A and lora_B but lora_embedding_A and lora_embedding_A. These are btw. nn.ParameterDicts, not nn.Linear like for linear layers. So the weight should be something like foobar.lora_embedding_A[adapter_name], without any .weight at the end.

@spezialspezial
Copy link
Contributor

Regarding parameters ending in .diff or .diff_b. I crawled all my local LoRAs and there they appear only in LoRAs from this guy with the RM_ prefix and are all completely zero.

I tried "Extract and Save LoRA" myself and found a bias_dif switch. With this checked, extracting a LoRA from a "finetuned" transformer (often just someone merging several LoRAs into the transformer and distributing that I suspect) I get a few nonzero .diff_b parameters. Pretty small values but not entirely zero. Didn't check extracting from text_encoder.

So in conclusion I think .dif is supposed to be lora_A.bias and .dif_b is lora_B.bias with a yet to be determined impact on generations and if they are important when extracting from an actually fully finetuned transformer.

@spezialspezial
Copy link
Contributor

The xlabs LoRA works for me too

@spezialspezial
Copy link
Contributor

Ah sorry, I think there is a regression with vanilla Kohya-LoRAs now.
https://civitai.com/models/305429?modelVersionId=763709
gives
ValueError: Invalid LoRA checkpoint.

@sayakpaul
Copy link
Member Author

@spezialspezial that should be fixed now. Would be nice if you could check it again.

@sayakpaul
Copy link
Member Author

@spezialspezial I have also pushed a couple of updates in 5c4976b to better filter T5 and other diff related keys. Please check it out and let me know.

@spezialspezial
Copy link
Contributor

spezialspezial commented Mar 7, 2025

The WizardWhitebeard LoRA (kind: kohya, dtype: bfloat16, rank: 20) works again

@spezialspezial
Copy link
Contributor

Not expecting you to fix it but check out this odd bird:

https://civitai.com/models/1269030?modelVersionId=1431428

lora_unet_double_blocks_0_img_attn_proj.alpha              shape: ()           val: 10.0
lora_unet_double_blocks_0_img_attn_proj.lora_down.weight   shape: (10, 3072)   val: [-0.0025, 0.014, 8.6e-05, ..]
lora_unet_double_blocks_0_img_attn_proj.lora_up.weight     shape: (3072, 10)   val: [0.06, -0.021, 0.05, ..]
lora_unet_double_blocks_0_img_attn_qkv.alpha               shape: ()           val: 10.0
lora_unet_double_blocks_0_img_attn_qkv.lora_down.weight    shape: (10, 3072)   val: [0.002, 0.061, -0.0046, ..]
lora_unet_double_blocks_0_img_attn_qkv.lora_up.weight      shape: (9216, 10)   val: [-0.045, 0.0053, -0.026, ..]
lora_unet_double_blocks_0_img_mlp_0.alpha                  shape: ()           val: 18.0
lora_unet_double_blocks_0_img_mlp_0.lora_down.weight       shape: (18, 3072)   val: [0.00076, 0.023, 0.0089, ..]
lora_unet_double_blocks_0_img_mlp_0.lora_up.weight         shape: (12288, 18)  val: [-0.021, -0.0025, 0.00024, ..]
lora_unet_double_blocks_0_img_mlp_2.alpha                  shape: ()           val: 11.0
lora_unet_double_blocks_0_img_mlp_2.lora_down.weight       shape: (11, 12288)  val: [0.02, -5.7e-05, -0.016, ..]
lora_unet_double_blocks_0_img_mlp_2.lora_up.weight         shape: (3072, 11)   val: [-0.073, -0.015, -0.06, ..]
lora_unet_double_blocks_0_img_mod_lin.alpha                shape: ()           val: 12.0
lora_unet_double_blocks_0_img_mod_lin.lora_down.weight     shape: (12, 3072)   val: [-0.0051, -0.012, -0.018, ..]
lora_unet_double_blocks_0_img_mod_lin.lora_up.weight       shape: (18432, 12)  val: [-0.0046, 0.021, -0.0069, ..]
lora_unet_double_blocks_0_txt_attn_proj.alpha              shape: ()           val: 3.0
lora_unet_double_blocks_0_txt_attn_proj.lora_down.weight   shape: (3, 3072)    val: [-0.012, 0.0021, -0.0025, ..]
lora_unet_double_blocks_0_txt_attn_proj.lora_up.weight     shape: (3072, 3)    val: [-0.031, 0.015, 0.0041, ..]
lora_unet_double_blocks_0_txt_attn_qkv.alpha               shape: ()           val: 4.0
lora_unet_double_blocks_0_txt_attn_qkv.lora_down.weight    shape: (4, 3072)    val: [0.0075, -0.017, -0.0094, ..]
lora_unet_double_blocks_0_txt_attn_qkv.lora_up.weight      shape: (9216, 4)    val: [-0.061, 0.034, 0.033, ..]
lora_unet_double_blocks_0_txt_mlp_0.alpha                  shape: ()           val: 2.0
lora_unet_double_blocks_0_txt_mlp_0.lora_down.weight       shape: (2, 3072)    val: [-0.012, 0.042, 0.013, ..]
lora_unet_double_blocks_0_txt_mlp_0.lora_up.weight         shape: (12288, 2)   val: [0.035, 0.045, -0.21, ..]
lora_unet_double_blocks_0_txt_mlp_2.alpha                  shape: ()           val: 4.0
lora_unet_double_blocks_0_txt_mlp_2.lora_down.weight       shape: (4, 12288)   val: [0.0022, 0.0015, 0.021, ..]
lora_unet_double_blocks_0_txt_mlp_2.lora_up.weight         shape: (3072, 4)    val: [0.34, -0.21, 0.13, ..]
lora_unet_double_blocks_0_txt_mod_lin.alpha                shape: ()           val: 2.0
lora_unet_double_blocks_0_txt_mod_lin.lora_down.weight     shape: (2, 3072)    val: [-0.023, 0.011, -0.015, ..]
lora_unet_double_blocks_0_txt_mod_lin.lora_up.weight       shape: (18432, 2)   val: [0.14, -0.016, -0.13, ..]

Pay attention to the smaller dimension of the tensors.

@spezialspezial
Copy link
Contributor

RandomMaxx Illustrify (kind: comfyui, dtype: float16, rank: 32) works

@sayakpaul
Copy link
Member Author

Yeah it has different ranks for different layers -- something we have supported for a long time. Any errors?

@spezialspezial
Copy link
Contributor

Yeah, fails for me with: size mismatch for transformer_blocks.17.attn.to_v.lora_B.Masha Babko.weight: copying a param with shape torch.Size([3072, 10]) from checkpoint, the shape in current model is torch.Size([3072, 8]).

@sayakpaul
Copy link
Member Author

Okay then for some reason, the rank pattern is not getting updated. Will take a look soon after the ICCV submissions. Hope that is okay.

@spezialspezial
Copy link
Contributor

You have a talk there? Nice! Is it about fighting sleeplessness from seeing too many obscure LoRAs?

@sayakpaul
Copy link
Member Author

sayakpaul commented Mar 8, 2025

@BenjaminBossan @hlky the _maybe_adjust_config() bug has struck us again I am afraid.

Code:

from diffusers import DiffusionPipeline 
import torch 

repo_id = "black-forest-labs/FLUX.1-dev"
lora_name = "babko.safetensors"
pipeline = DiffusionPipeline.from_pretrained(repo_id, torch_dtype=torch.bfloat16)
pipeline.load_lora_weights(
    "sayakpaul/different-lora-from-civitai", weight_name=lora_name
)

Consider the following two keys:

transformer_blocks.0.attn.to_k
single_transformer_blocks.0.attn.to_k

This becomes ambiguous for the adjustment method even after #10808 and #11004. I am running out of sane ways to handle this, hence the mention here. Maybe the PEFT PR here could be prioritized.

The LoRA tested in #10808 didn't have the issue because it has the same rank for all the layers unlike the current LoRA we're testing.

This is what is happening to the config before and after the adjustment:

Before
{
    "r": 4,
    "lora_alpha": 10,
    "rank_pattern": {
        "transformer_blocks.0.attn.to_out.0": 10,
        "transformer_blocks.0.attn.to_q": 10,
        "transformer_blocks.0.attn.to_k": 10,
        "transformer_blocks.0.attn.to_v": 10,
        "transformer_blocks.0.ff.net.0.proj": 18,
        "transformer_blocks.0.ff.net.2": 11,
        "transformer_blocks.0.norm1.linear": 12,
        "transformer_blocks.0.attn.to_add_out": 3,
        "transformer_blocks.0.ff_context.net.0.proj": 2,
        "transformer_blocks.0.norm1_context.linear": 2,
        "transformer_blocks.1.attn.to_out.0": 10,
        "transformer_blocks.1.attn.to_q": 9,
        "transformer_blocks.1.attn.to_k": 9,
        "transformer_blocks.1.attn.to_v": 9,
        "transformer_blocks.1.ff.net.0.proj": 29,
        "transformer_blocks.1.ff.net.2": 9,
        "transformer_blocks.1.norm1.linear": 20,
        "transformer_blocks.1.attn.to_add_out": 3,
        "transformer_blocks.1.attn.add_q_proj": 3,
        "transformer_blocks.1.attn.add_k_proj": 3,
        "transformer_blocks.1.attn.add_v_proj": 3,
        "transformer_blocks.1.ff_context.net.0.proj": 3,
        "transformer_blocks.1.ff_context.net.2": 2,
        "transformer_blocks.1.norm1_context.linear": 2,
        "transformer_blocks.2.attn.to_out.0": 13,
        "transformer_blocks.2.attn.to_q": 17,
        "transformer_blocks.2.attn.to_k": 17,
        "transformer_blocks.2.attn.to_v": 17,
        "transformer_blocks.2.ff.net.0.proj": 26,
        "transformer_blocks.2.ff.net.2": 8,
        "transformer_blocks.2.norm1.linear": 20,
        "transformer_blocks.2.attn.to_add_out": 5,
        "transformer_blocks.3.attn.to_out.0": 13,
        "transformer_blocks.3.attn.to_q": 13,
        "transformer_blocks.3.attn.to_k": 13,
        "transformer_blocks.3.attn.to_v": 13,
        "transformer_blocks.3.ff.net.0.proj": 11,
        "transformer_blocks.3.ff.net.2": 5,
        "transformer_blocks.3.norm1.linear": 10,
        "transformer_blocks.3.attn.to_add_out": 5,
        "transformer_blocks.3.attn.add_q_proj": 7,
        "transformer_blocks.3.attn.add_k_proj": 7,
        "transformer_blocks.3.attn.add_v_proj": 7,
        "transformer_blocks.3.ff_context.net.0.proj": 3,
        "transformer_blocks.3.ff_context.net.2": 5,
        "transformer_blocks.3.norm1_context.linear": 5,
        "transformer_blocks.4.attn.to_out.0": 8,
        "transformer_blocks.4.attn.to_q": 16,
        "transformer_blocks.4.attn.to_k": 16,
        "transformer_blocks.4.attn.to_v": 16,
        "transformer_blocks.4.ff.net.0.proj": 8,
        "transformer_blocks.4.ff.net.2": 6,
        "transformer_blocks.4.norm1.linear": 8,
        "transformer_blocks.4.attn.to_add_out": 5,
        "transformer_blocks.4.attn.add_q_proj": 6,
        "transformer_blocks.4.attn.add_k_proj": 6,
        "transformer_blocks.4.attn.add_v_proj": 6,
        "transformer_blocks.4.ff_context.net.0.proj": 3,
        "transformer_blocks.4.norm1_context.linear": 5,
        "transformer_blocks.5.attn.to_out.0": 9,
        "transformer_blocks.5.attn.to_q": 8,
        "transformer_blocks.5.attn.to_k": 8,
        "transformer_blocks.5.attn.to_v": 8,
        "transformer_blocks.5.ff.net.0.proj": 10,
        "transformer_blocks.5.ff.net.2": 7,
        "transformer_blocks.5.norm1.linear": 8,
        "transformer_blocks.5.attn.add_q_proj": 3,
        "transformer_blocks.5.attn.add_k_proj": 3,
        "transformer_blocks.5.attn.add_v_proj": 3,
        "transformer_blocks.5.ff_context.net.2": 2,
        "transformer_blocks.5.norm1_context.linear": 3,
        "transformer_blocks.6.attn.to_out.0": 11,
        "transformer_blocks.6.attn.to_q": 11,
        "transformer_blocks.6.attn.to_k": 11,
        "transformer_blocks.6.attn.to_v": 11,
        "transformer_blocks.6.ff.net.0.proj": 11,
        "transformer_blocks.6.ff.net.2": 5,
        "transformer_blocks.6.norm1.linear": 8,
        "transformer_blocks.6.attn.to_add_out": 5,
        "transformer_blocks.6.attn.add_q_proj": 3,
        "transformer_blocks.6.attn.add_k_proj": 3,
        "transformer_blocks.6.attn.add_v_proj": 3,
        "transformer_blocks.6.ff_context.net.0.proj": 5,
        "transformer_blocks.6.norm1_context.linear": 2,
        "transformer_blocks.7.attn.to_out.0": 12,
        "transformer_blocks.7.attn.to_q": 10,
        "transformer_blocks.7.attn.to_k": 10,
        "transformer_blocks.7.attn.to_v": 10,
        "transformer_blocks.7.ff.net.0.proj": 12,
        "transformer_blocks.7.ff.net.2": 5,
        "transformer_blocks.7.norm1.linear": 8,
        "transformer_blocks.7.attn.add_q_proj": 5,
        "transformer_blocks.7.attn.add_k_proj": 5,
        "transformer_blocks.7.attn.add_v_proj": 5,
        "transformer_blocks.7.ff_context.net.0.proj": 3,
        "transformer_blocks.7.norm1_context.linear": 6,
        "transformer_blocks.8.attn.to_out.0": 13,
        "transformer_blocks.8.attn.to_q": 13,
        "transformer_blocks.8.attn.to_k": 13,
        "transformer_blocks.8.attn.to_v": 13,
        "transformer_blocks.8.ff.net.0.proj": 9,
        "transformer_blocks.8.ff.net.2": 6,
        "transformer_blocks.8.norm1.linear": 7,
        "transformer_blocks.8.attn.to_add_out": 7,
        "transformer_blocks.8.ff_context.net.2": 6,
        "transformer_blocks.8.norm1_context.linear": 3,
        "transformer_blocks.9.attn.to_out.0": 7,
        "transformer_blocks.9.attn.to_q": 10,
        "transformer_blocks.9.attn.to_k": 10,
        "transformer_blocks.9.attn.to_v": 10,
        "transformer_blocks.9.ff.net.0.proj": 12,
        "transformer_blocks.9.ff.net.2": 6,
        "transformer_blocks.9.norm1.linear": 8,
        "transformer_blocks.9.attn.to_add_out": 7,
        "transformer_blocks.9.attn.add_q_proj": 6,
        "transformer_blocks.9.attn.add_k_proj": 6,
        "transformer_blocks.9.attn.add_v_proj": 6,
        "transformer_blocks.9.ff_context.net.0.proj": 3,
        "transformer_blocks.9.ff_context.net.2": 5,
        "transformer_blocks.10.attn.to_out.0": 6,
        "transformer_blocks.10.attn.to_q": 7,
        "transformer_blocks.10.attn.to_k": 7,
        "transformer_blocks.10.attn.to_v": 7,
        "transformer_blocks.10.ff.net.0.proj": 8,
        "transformer_blocks.10.ff.net.2": 5,
        "transformer_blocks.10.norm1.linear": 7,
        "transformer_blocks.10.attn.to_add_out": 6,
        "transformer_blocks.10.attn.add_q_proj": 7,
        "transformer_blocks.10.attn.add_k_proj": 7,
        "transformer_blocks.10.attn.add_v_proj": 7,
        "transformer_blocks.10.ff_context.net.0.proj": 3,
        "transformer_blocks.10.norm1_context.linear": 7,
        "transformer_blocks.11.attn.to_out.0": 9,
        "transformer_blocks.11.attn.to_q": 9,
        "transformer_blocks.11.attn.to_k": 9,
        "transformer_blocks.11.attn.to_v": 9,
        "transformer_blocks.11.ff.net.0.proj": 10,
        "transformer_blocks.11.ff.net.2": 6,
        "transformer_blocks.11.norm1.linear": 7,
        "transformer_blocks.11.attn.to_add_out": 3,
        "transformer_blocks.11.attn.add_q_proj": 6,
        "transformer_blocks.11.attn.add_k_proj": 6,
        "transformer_blocks.11.attn.add_v_proj": 6,
        "transformer_blocks.11.ff_context.net.0.proj": 3,
        "transformer_blocks.11.ff_context.net.2": 5,
        "transformer_blocks.12.attn.to_out.0": 11,
        "transformer_blocks.12.attn.to_q": 15,
        "transformer_blocks.12.attn.to_k": 15,
        "transformer_blocks.12.attn.to_v": 15,
        "transformer_blocks.12.ff.net.0.proj": 12,
        "transformer_blocks.12.ff.net.2": 6,
        "transformer_blocks.12.norm1.linear": 7,
        "transformer_blocks.12.attn.to_add_out": 5,
        "transformer_blocks.12.attn.add_q_proj": 5,
        "transformer_blocks.12.attn.add_k_proj": 5,
        "transformer_blocks.12.attn.add_v_proj": 5,
        "transformer_blocks.12.ff_context.net.0.proj": 3,
        "transformer_blocks.12.norm1_context.linear": 5,
        "transformer_blocks.13.attn.to_out.0": 8,
        "transformer_blocks.13.attn.to_q": 8,
        "transformer_blocks.13.attn.to_k": 8,
        "transformer_blocks.13.attn.to_v": 8,
        "transformer_blocks.13.ff.net.0.proj": 9,
        "transformer_blocks.13.norm1.linear": 5,
        "transformer_blocks.13.attn.to_add_out": 5,
        "transformer_blocks.13.ff_context.net.0.proj": 2,
        "transformer_blocks.13.ff_context.net.2": 3,
        "transformer_blocks.13.norm1_context.linear": 6,
        "transformer_blocks.14.attn.to_out.0": 11,
        "transformer_blocks.14.attn.to_q": 10,
        "transformer_blocks.14.attn.to_k": 10,
        "transformer_blocks.14.attn.to_v": 10,
        "transformer_blocks.14.ff.net.0.proj": 12,
        "transformer_blocks.14.ff.net.2": 6,
        "transformer_blocks.14.norm1.linear": 5,
        "transformer_blocks.14.attn.to_add_out": 3,
        "transformer_blocks.14.attn.add_q_proj": 5,
        "transformer_blocks.14.attn.add_k_proj": 5,
        "transformer_blocks.14.attn.add_v_proj": 5,
        "transformer_blocks.14.ff_context.net.0.proj": 3,
        "transformer_blocks.14.norm1_context.linear": 3,
        "transformer_blocks.15.attn.to_out.0": 9,
        "transformer_blocks.15.attn.to_q": 8,
        "transformer_blocks.15.attn.to_k": 8,
        "transformer_blocks.15.attn.to_v": 8,
        "transformer_blocks.15.ff.net.0.proj": 12,
        "transformer_blocks.15.ff.net.2": 7,
        "transformer_blocks.15.norm1.linear": 6,
        "transformer_blocks.15.attn.add_q_proj": 3,
        "transformer_blocks.15.attn.add_k_proj": 3,
        "transformer_blocks.15.attn.add_v_proj": 3,
        "transformer_blocks.16.attn.to_out.0": 10,
        "transformer_blocks.16.attn.to_q": 13,
        "transformer_blocks.16.attn.to_k": 13,
        "transformer_blocks.16.attn.to_v": 13,
        "transformer_blocks.16.ff.net.0.proj": 12,
        "transformer_blocks.16.ff.net.2": 7,
        "transformer_blocks.16.norm1.linear": 6,
        "transformer_blocks.16.attn.add_q_proj": 6,
        "transformer_blocks.16.attn.add_k_proj": 6,
        "transformer_blocks.16.attn.add_v_proj": 6,
        "transformer_blocks.16.ff_context.net.2": 5,
        "transformer_blocks.16.norm1_context.linear": 3,
        "transformer_blocks.17.attn.to_out.0": 8,
        "transformer_blocks.17.attn.to_q": 10,
        "transformer_blocks.17.attn.to_k": 10,
        "transformer_blocks.17.attn.to_v": 10,
        "transformer_blocks.17.ff.net.0.proj": 9,
        "transformer_blocks.17.ff.net.2": 9,
        "transformer_blocks.17.norm1.linear": 8,
        "transformer_blocks.17.attn.to_add_out": 5,
        "transformer_blocks.17.attn.add_q_proj": 5,
        "transformer_blocks.17.attn.add_k_proj": 5,
        "transformer_blocks.17.attn.add_v_proj": 5,
        "transformer_blocks.17.ff_context.net.0.proj": 2,
        "transformer_blocks.17.norm1_context.linear": 2,
        "transformer_blocks.18.attn.to_out.0": 5,
        "transformer_blocks.18.attn.to_q": 8,
        "transformer_blocks.18.attn.to_k": 8,
        "transformer_blocks.18.attn.to_v": 8,
        "transformer_blocks.18.ff.net.0.proj": 11,
        "transformer_blocks.18.ff.net.2": 8,
        "transformer_blocks.18.norm1.linear": 6,
        "transformer_blocks.18.ff_context.net.0.proj": 3,
        "transformer_blocks.18.norm1_context.linear": 7,
        "single_transformer_blocks.0.attn.to_q": 18,
        "single_transformer_blocks.0.attn.to_k": 18,
        "single_transformer_blocks.0.attn.to_v": 18,
        "single_transformer_blocks.0.proj_mlp": 18,
        "single_transformer_blocks.0.proj_out": 8,
        "single_transformer_blocks.0.norm.linear": 7,
        "single_transformer_blocks.1.attn.to_q": 22,
        "single_transformer_blocks.1.attn.to_k": 22,
        "single_transformer_blocks.1.attn.to_v": 22,
        "single_transformer_blocks.1.proj_mlp": 22,
        "single_transformer_blocks.1.proj_out": 7,
        "single_transformer_blocks.1.norm.linear": 7,
        "single_transformer_blocks.2.attn.to_q": 18,
        "single_transformer_blocks.2.attn.to_k": 18,
        "single_transformer_blocks.2.attn.to_v": 18,
        "single_transformer_blocks.2.proj_mlp": 18,
        "single_transformer_blocks.2.proj_out": 7,
        "single_transformer_blocks.2.norm.linear": 7,
        "single_transformer_blocks.3.attn.to_q": 21,
        "single_transformer_blocks.3.attn.to_k": 21,
        "single_transformer_blocks.3.attn.to_v": 21,
        "single_transformer_blocks.3.proj_mlp": 21,
        "single_transformer_blocks.3.proj_out": 7,
        "single_transformer_blocks.3.norm.linear": 8,
        "single_transformer_blocks.4.attn.to_q": 18,
        "single_transformer_blocks.4.attn.to_k": 18,
        "single_transformer_blocks.4.attn.to_v": 18,
        "single_transformer_blocks.4.proj_mlp": 18,
        "single_transformer_blocks.4.proj_out": 8,
        "single_transformer_blocks.4.norm.linear": 8,
        "single_transformer_blocks.5.attn.to_q": 20,
        "single_transformer_blocks.5.attn.to_k": 20,
        "single_transformer_blocks.5.attn.to_v": 20,
        "single_transformer_blocks.5.proj_mlp": 20,
        "single_transformer_blocks.5.proj_out": 8,
        "single_transformer_blocks.5.norm.linear": 9,
        "single_transformer_blocks.6.attn.to_q": 20,
        "single_transformer_blocks.6.attn.to_k": 20,
        "single_transformer_blocks.6.attn.to_v": 20,
        "single_transformer_blocks.6.proj_mlp": 20,
        "single_transformer_blocks.6.proj_out": 6,
        "single_transformer_blocks.6.norm.linear": 10,
        "single_transformer_blocks.7.attn.to_q": 21,
        "single_transformer_blocks.7.attn.to_k": 21,
        "single_transformer_blocks.7.attn.to_v": 21,
        "single_transformer_blocks.7.proj_mlp": 21,
        "single_transformer_blocks.7.proj_out": 8,
        "single_transformer_blocks.7.norm.linear": 10,
        "single_transformer_blocks.8.attn.to_q": 19,
        "single_transformer_blocks.8.attn.to_k": 19,
        "single_transformer_blocks.8.attn.to_v": 19,
        "single_transformer_blocks.8.proj_mlp": 19,
        "single_transformer_blocks.8.proj_out": 7,
        "single_transformer_blocks.8.norm.linear": 7,
        "single_transformer_blocks.9.attn.to_q": 22,
        "single_transformer_blocks.9.attn.to_k": 22,
        "single_transformer_blocks.9.attn.to_v": 22,
        "single_transformer_blocks.9.proj_mlp": 22,
        "single_transformer_blocks.9.proj_out": 9,
        "single_transformer_blocks.9.norm.linear": 8,
        "single_transformer_blocks.10.attn.to_q": 22,
        "single_transformer_blocks.10.attn.to_k": 22,
        "single_transformer_blocks.10.attn.to_v": 22,
        "single_transformer_blocks.10.proj_mlp": 22,
        "single_transformer_blocks.10.proj_out": 11,
        "single_transformer_blocks.10.norm.linear": 11,
        "single_transformer_blocks.11.attn.to_q": 22,
        "single_transformer_blocks.11.attn.to_k": 22,
        "single_transformer_blocks.11.attn.to_v": 22,
        "single_transformer_blocks.11.proj_mlp": 22,
        "single_transformer_blocks.11.proj_out": 10,
        "single_transformer_blocks.11.norm.linear": 10,
        "single_transformer_blocks.12.attn.to_q": 23,
        "single_transformer_blocks.12.attn.to_k": 23,
        "single_transformer_blocks.12.attn.to_v": 23,
        "single_transformer_blocks.12.proj_mlp": 23,
        "single_transformer_blocks.12.proj_out": 10,
        "single_transformer_blocks.12.norm.linear": 8,
        "single_transformer_blocks.13.attn.to_q": 22,
        "single_transformer_blocks.13.attn.to_k": 22,
        "single_transformer_blocks.13.attn.to_v": 22,
        "single_transformer_blocks.13.proj_mlp": 22,
        "single_transformer_blocks.13.proj_out": 10,
        "single_transformer_blocks.13.norm.linear": 12,
        "single_transformer_blocks.14.attn.to_q": 24,
        "single_transformer_blocks.14.attn.to_k": 24,
        "single_transformer_blocks.14.attn.to_v": 24,
        "single_transformer_blocks.14.proj_mlp": 24,
        "single_transformer_blocks.14.proj_out": 11,
        "single_transformer_blocks.14.norm.linear": 11,
        "single_transformer_blocks.15.attn.to_q": 25,
        "single_transformer_blocks.15.attn.to_k": 25,
        "single_transformer_blocks.15.attn.to_v": 25,
        "single_transformer_blocks.15.proj_mlp": 25,
        "single_transformer_blocks.15.proj_out": 10,
        "single_transformer_blocks.15.norm.linear": 11,
        "single_transformer_blocks.16.attn.to_q": 26,
        "single_transformer_blocks.16.attn.to_k": 26,
        "single_transformer_blocks.16.attn.to_v": 26,
        "single_transformer_blocks.16.proj_mlp": 26,
        "single_transformer_blocks.16.proj_out": 10,
        "single_transformer_blocks.16.norm.linear": 9,
        "single_transformer_blocks.17.attn.to_q": 28,
        "single_transformer_blocks.17.attn.to_k": 28,
        "single_transformer_blocks.17.attn.to_v": 28,
        "single_transformer_blocks.17.proj_mlp": 28,
        "single_transformer_blocks.17.proj_out": 10,
        "single_transformer_blocks.17.norm.linear": 9,
        "single_transformer_blocks.18.attn.to_q": 28,
        "single_transformer_blocks.18.attn.to_k": 28,
        "single_transformer_blocks.18.attn.to_v": 28,
        "single_transformer_blocks.18.proj_mlp": 28,
        "single_transformer_blocks.18.proj_out": 8,
        "single_transformer_blocks.18.norm.linear": 11,
        "single_transformer_blocks.19.attn.to_q": 27,
        "single_transformer_blocks.19.attn.to_k": 27,
        "single_transformer_blocks.19.attn.to_v": 27,
        "single_transformer_blocks.19.proj_mlp": 27,
        "single_transformer_blocks.19.proj_out": 10,
        "single_transformer_blocks.19.norm.linear": 10,
        "single_transformer_blocks.20.attn.to_q": 29,
        "single_transformer_blocks.20.attn.to_k": 29,
        "single_transformer_blocks.20.attn.to_v": 29,
        "single_transformer_blocks.20.proj_mlp": 29,
        "single_transformer_blocks.20.proj_out": 11,
        "single_transformer_blocks.20.norm.linear": 12,
        "single_transformer_blocks.21.attn.to_q": 26,
        "single_transformer_blocks.21.attn.to_k": 26,
        "single_transformer_blocks.21.attn.to_v": 26,
        "single_transformer_blocks.21.proj_mlp": 26,
        "single_transformer_blocks.21.proj_out": 11,
        "single_transformer_blocks.21.norm.linear": 11,
        "single_transformer_blocks.22.attn.to_q": 26,
        "single_transformer_blocks.22.attn.to_k": 26,
        "single_transformer_blocks.22.attn.to_v": 26,
        "single_transformer_blocks.22.proj_mlp": 26,
        "single_transformer_blocks.22.proj_out": 11,
        "single_transformer_blocks.22.norm.linear": 10,
        "single_transformer_blocks.23.attn.to_q": 30,
        "single_transformer_blocks.23.attn.to_k": 30,
        "single_transformer_blocks.23.attn.to_v": 30,
        "single_transformer_blocks.23.proj_mlp": 30,
        "single_transformer_blocks.23.proj_out": 11,
        "single_transformer_blocks.23.norm.linear": 11,
        "single_transformer_blocks.24.attn.to_q": 28,
        "single_transformer_blocks.24.attn.to_k": 28,
        "single_transformer_blocks.24.attn.to_v": 28,
        "single_transformer_blocks.24.proj_mlp": 28,
        "single_transformer_blocks.24.proj_out": 12,
        "single_transformer_blocks.24.norm.linear": 11,
        "single_transformer_blocks.25.attn.to_q": 28,
        "single_transformer_blocks.25.attn.to_k": 28,
        "single_transformer_blocks.25.attn.to_v": 28,
        "single_transformer_blocks.25.proj_mlp": 28,
        "single_transformer_blocks.25.proj_out": 14,
        "single_transformer_blocks.25.norm.linear": 9,
        "single_transformer_blocks.26.attn.to_q": 28,
        "single_transformer_blocks.26.attn.to_k": 28,
        "single_transformer_blocks.26.attn.to_v": 28,
        "single_transformer_blocks.26.proj_mlp": 28,
        "single_transformer_blocks.26.proj_out": 13,
        "single_transformer_blocks.26.norm.linear": 9,
        "single_transformer_blocks.27.attn.to_q": 29,
        "single_transformer_blocks.27.attn.to_k": 29,
        "single_transformer_blocks.27.attn.to_v": 29,
        "single_transformer_blocks.27.proj_mlp": 29,
        "single_transformer_blocks.27.proj_out": 14,
        "single_transformer_blocks.27.norm.linear": 8,
        "single_transformer_blocks.28.attn.to_q": 29,
        "single_transformer_blocks.28.attn.to_k": 29,
        "single_transformer_blocks.28.attn.to_v": 29,
        "single_transformer_blocks.28.proj_mlp": 29,
        "single_transformer_blocks.28.proj_out": 14,
        "single_transformer_blocks.28.norm.linear": 12,
        "single_transformer_blocks.29.attn.to_q": 30,
        "single_transformer_blocks.29.attn.to_k": 30,
        "single_transformer_blocks.29.attn.to_v": 30,
        "single_transformer_blocks.29.proj_mlp": 30,
        "single_transformer_blocks.29.proj_out": 14,
        "single_transformer_blocks.29.norm.linear": 10,
        "single_transformer_blocks.30.attn.to_q": 32,
        "single_transformer_blocks.30.attn.to_k": 32,
        "single_transformer_blocks.30.attn.to_v": 32,
        "single_transformer_blocks.30.proj_mlp": 32,
        "single_transformer_blocks.30.proj_out": 15,
        "single_transformer_blocks.30.norm.linear": 9,
        "single_transformer_blocks.31.attn.to_q": 30,
        "single_transformer_blocks.31.attn.to_k": 30,
        "single_transformer_blocks.31.attn.to_v": 30,
        "single_transformer_blocks.31.proj_mlp": 30,
        "single_transformer_blocks.31.proj_out": 16,
        "single_transformer_blocks.31.norm.linear": 11,
        "single_transformer_blocks.32.attn.to_q": 30,
        "single_transformer_blocks.32.attn.to_k": 30,
        "single_transformer_blocks.32.attn.to_v": 30,
        "single_transformer_blocks.32.proj_mlp": 30,
        "single_transformer_blocks.32.proj_out": 15,
        "single_transformer_blocks.32.norm.linear": 9,
        "single_transformer_blocks.33.attn.to_q": 32,
        "single_transformer_blocks.33.attn.to_k": 32,
        "single_transformer_blocks.33.attn.to_v": 32,
        "single_transformer_blocks.33.proj_mlp": 32,
        "single_transformer_blocks.33.proj_out": 16,
        "single_transformer_blocks.33.norm.linear": 8,
        "single_transformer_blocks.34.attn.to_q": 30,
        "single_transformer_blocks.34.attn.to_k": 30,
        "single_transformer_blocks.34.attn.to_v": 30,
        "single_transformer_blocks.34.proj_mlp": 30,
        "single_transformer_blocks.34.proj_out": 16,
        "single_transformer_blocks.34.norm.linear": 7,
        "single_transformer_blocks.35.attn.to_q": 31,
        "single_transformer_blocks.35.attn.to_k": 31,
        "single_transformer_blocks.35.attn.to_v": 31,
        "single_transformer_blocks.35.proj_mlp": 31,
        "single_transformer_blocks.35.proj_out": 14,
        "single_transformer_blocks.35.norm.linear": 8,
        "single_transformer_blocks.36.attn.to_q": 30,
        "single_transformer_blocks.36.attn.to_k": 30,
        "single_transformer_blocks.36.attn.to_v": 30,
        "single_transformer_blocks.36.proj_mlp": 30,
        "single_transformer_blocks.36.proj_out": 12,
        "single_transformer_blocks.36.norm.linear": 9,
        "single_transformer_blocks.37.attn.to_q": 17,
        "single_transformer_blocks.37.attn.to_k": 17,
        "single_transformer_blocks.37.attn.to_v": 17,
        "single_transformer_blocks.37.proj_mlp": 17,
        "single_transformer_blocks.37.proj_out": 7,
        "single_transformer_blocks.37.norm.linear": 9
    },
    "alpha_pattern": {},
    "target_modules": [
        "transformer_blocks.11.ff_context.net.2",
        "single_transformer_blocks.23.attn.to_k",
        "transformer_blocks.8.attn.add_v_proj",
        "single_transformer_blocks.5.proj_out",
        "transformer_blocks.18.attn.to_add_out",
        "single_transformer_blocks.33.attn.to_v",
        "transformer_blocks.17.norm1.linear",
        "single_transformer_blocks.1.proj_mlp",
        "transformer_blocks.4.attn.to_v",
        "transformer_blocks.7.attn.to_v",
        "transformer_blocks.13.attn.to_add_out",
        "transformer_blocks.10.ff.net.2",
        "single_transformer_blocks.0.attn.to_v",
        "single_transformer_blocks.22.attn.to_v",
        "transformer_blocks.6.ff_context.net.2",
        "transformer_blocks.15.attn.to_q",
        "single_transformer_blocks.25.proj_mlp",
        "single_transformer_blocks.17.attn.to_q",
        "single_transformer_blocks.26.proj_out",
        "single_transformer_blocks.30.attn.to_k",
        "single_transformer_blocks.32.attn.to_v",
        "transformer_blocks.1.ff.net.2",
        "single_transformer_blocks.10.proj_out",
        "single_transformer_blocks.14.norm.linear",
        "transformer_blocks.15.attn.to_out.0",
        "transformer_blocks.17.ff.net.2",
        "single_transformer_blocks.34.attn.to_v",
        "single_transformer_blocks.4.proj_mlp",
        "transformer_blocks.7.attn.to_q",
        "transformer_blocks.0.ff.net.0.proj",
        "transformer_blocks.14.attn.add_v_proj",
        "transformer_blocks.8.ff_context.net.0.proj",
        "single_transformer_blocks.28.attn.to_q",
        "transformer_blocks.13.attn.add_k_proj",
        "single_transformer_blocks.2.proj_mlp",
        "single_transformer_blocks.21.attn.to_v",
        "single_transformer_blocks.9.attn.to_k",
        "single_transformer_blocks.9.attn.to_v",
        "transformer_blocks.17.attn.to_v",
        "single_transformer_blocks.31.attn.to_k",
        "transformer_blocks.10.norm1_context.linear",
        "transformer_blocks.17.attn.add_q_proj",
        "transformer_blocks.6.ff.net.2",
        "transformer_blocks.12.ff.net.0.proj",
        "single_transformer_blocks.1.attn.to_k",
        "transformer_blocks.1.attn.to_v",
        "single_transformer_blocks.23.attn.to_q",
        "transformer_blocks.13.ff.net.0.proj",
        "single_transformer_blocks.17.proj_mlp",
        "transformer_blocks.8.attn.add_k_proj",
        "transformer_blocks.5.ff_context.net.0.proj",
        "single_transformer_blocks.5.norm.linear",
        "transformer_blocks.15.attn.to_k",
        "single_transformer_blocks.20.attn.to_v",
        "single_transformer_blocks.36.attn.to_q",
        "transformer_blocks.9.norm1.linear",
        "single_transformer_blocks.5.attn.to_q",
        "single_transformer_blocks.10.attn.to_v",
        "single_transformer_blocks.30.attn.to_q",
        "transformer_blocks.12.attn.to_add_out",
        "transformer_blocks.12.ff_context.net.2",
        "transformer_blocks.11.ff_context.net.0.proj",
        "transformer_blocks.2.ff_context.net.0.proj",
        "single_transformer_blocks.6.attn.to_q",
        "single_transformer_blocks.3.attn.to_k",
        "single_transformer_blocks.13.attn.to_k",
        "single_transformer_blocks.16.attn.to_q",
        "single_transformer_blocks.27.attn.to_q",
        "transformer_blocks.7.ff_context.net.0.proj",
        "single_transformer_blocks.23.proj_out",
        "transformer_blocks.12.attn.add_k_proj",
        "single_transformer_blocks.15.attn.to_q",
        "single_transformer_blocks.37.proj_out",
        "transformer_blocks.3.attn.to_v",
        "transformer_blocks.17.attn.add_v_proj",
        "transformer_blocks.2.norm1_context.linear",
        "single_transformer_blocks.32.norm.linear",
        "transformer_blocks.12.ff.net.2",
        "single_transformer_blocks.20.attn.to_k",
        "transformer_blocks.2.ff.net.2",
        "single_transformer_blocks.3.proj_out",
        "transformer_blocks.18.attn.to_q",
        "single_transformer_blocks.14.attn.to_q",
        "single_transformer_blocks.11.proj_mlp",
        "single_transformer_blocks.14.attn.to_v",
        "transformer_blocks.13.attn.add_q_proj",
        "transformer_blocks.15.ff_context.net.0.proj",
        "single_transformer_blocks.8.proj_mlp",
        "single_transformer_blocks.16.attn.to_k",
        "single_transformer_blocks.27.norm.linear",
        "transformer_blocks.8.ff.net.0.proj",
        "single_transformer_blocks.33.norm.linear",
        "transformer_blocks.10.attn.to_add_out",
        "single_transformer_blocks.33.proj_out",
        "transformer_blocks.16.attn.add_v_proj",
        "transformer_blocks.12.attn.to_out.0",
        "transformer_blocks.15.attn.add_q_proj",
        "transformer_blocks.2.attn.to_v",
        "single_transformer_blocks.17.proj_out",
        "transformer_blocks.5.attn.add_k_proj",
        "transformer_blocks.12.attn.to_q",
        "single_transformer_blocks.0.attn.to_q",
        "transformer_blocks.1.attn.add_q_proj",
        "transformer_blocks.14.attn.to_out.0",
        "single_transformer_blocks.33.attn.to_k",
        "transformer_blocks.3.ff.net.0.proj",
        "single_transformer_blocks.18.proj_mlp",
        "single_transformer_blocks.26.attn.to_k",
        "transformer_blocks.4.attn.add_q_proj",
        "transformer_blocks.10.attn.to_out.0",
        "transformer_blocks.13.ff_context.net.2",
        "transformer_blocks.6.ff.net.0.proj",
        "transformer_blocks.2.attn.to_q",
        "transformer_blocks.17.attn.to_add_out",
        "single_transformer_blocks.24.attn.to_q",
        "single_transformer_blocks.32.attn.to_q",
        "transformer_blocks.16.ff.net.2",
        "single_transformer_blocks.12.norm.linear",
        "transformer_blocks.4.attn.to_add_out",
        "single_transformer_blocks.28.attn.to_k",
        "transformer_blocks.18.ff_context.net.2",
        "transformer_blocks.9.attn.to_q",
        "transformer_blocks.8.norm1.linear",
        "transformer_blocks.16.attn.to_k",
        "transformer_blocks.13.attn.to_q",
        "transformer_blocks.5.attn.to_add_out",
        "transformer_blocks.7.attn.to_add_out",
        "transformer_blocks.9.attn.add_v_proj",
        "transformer_blocks.1.attn.to_add_out",
        "transformer_blocks.1.ff_context.net.0.proj",
        "transformer_blocks.2.ff.net.0.proj",
        "transformer_blocks.7.ff.net.2",
        "transformer_blocks.13.attn.add_v_proj",
        "single_transformer_blocks.35.attn.to_k",
        "single_transformer_blocks.29.norm.linear",
        "transformer_blocks.11.attn.to_out.0",
        "transformer_blocks.8.ff_context.net.2",
        "transformer_blocks.12.norm1.linear",
        "transformer_blocks.7.norm1.linear",
        "single_transformer_blocks.9.attn.to_q",
        "transformer_blocks.9.attn.add_k_proj",
        "single_transformer_blocks.14.attn.to_k",
        "single_transformer_blocks.3.attn.to_v",
        "single_transformer_blocks.37.norm.linear",
        "transformer_blocks.13.attn.to_v",
        "single_transformer_blocks.0.proj_mlp",
        "transformer_blocks.0.attn.to_q",
        "transformer_blocks.6.attn.to_add_out",
        "transformer_blocks.12.attn.to_v",
        "single_transformer_blocks.4.proj_out",
        "single_transformer_blocks.24.attn.to_v",
        "single_transformer_blocks.4.attn.to_v",
        "transformer_blocks.3.attn.to_k",
        "transformer_blocks.5.ff_context.net.2",
        "single_transformer_blocks.15.proj_out",
        "transformer_blocks.14.ff.net.2",
        "transformer_blocks.0.attn.add_v_proj",
        "single_transformer_blocks.15.norm.linear",
        "single_transformer_blocks.19.attn.to_q",
        "single_transformer_blocks.35.norm.linear",
        "transformer_blocks.11.attn.to_q",
        "single_transformer_blocks.6.proj_mlp",
        "transformer_blocks.16.attn.to_add_out",
        "single_transformer_blocks.14.proj_mlp",
        "transformer_blocks.10.attn.add_v_proj",
        "transformer_blocks.4.ff_context.net.2",
        "transformer_blocks.15.attn.to_add_out",
        "single_transformer_blocks.27.attn.to_v",
        "transformer_blocks.2.attn.to_k",
        "transformer_blocks.7.attn.add_q_proj",
        "single_transformer_blocks.1.norm.linear",
        "single_transformer_blocks.24.attn.to_k",
        "transformer_blocks.18.attn.add_q_proj",
        "transformer_blocks.14.ff_context.net.0.proj",
        "single_transformer_blocks.6.attn.to_k",
        "single_transformer_blocks.37.attn.to_v",
        "transformer_blocks.0.attn.add_q_proj",
        "transformer_blocks.0.attn.to_out.0",
        "single_transformer_blocks.5.attn.to_v",
        "single_transformer_blocks.5.attn.to_k",
        "transformer_blocks.18.norm1_context.linear",
        "transformer_blocks.4.ff.net.2",
        "single_transformer_blocks.29.attn.to_k",
        "transformer_blocks.6.ff_context.net.0.proj",
        "transformer_blocks.9.attn.to_k",
        "single_transformer_blocks.12.attn.to_q",
        "single_transformer_blocks.37.attn.to_k",
        "single_transformer_blocks.22.proj_mlp",
        "transformer_blocks.15.attn.add_v_proj",
        "transformer_blocks.4.norm1_context.linear",
        "single_transformer_blocks.13.norm.linear",
        "single_transformer_blocks.36.norm.linear",
        "transformer_blocks.5.norm1.linear",
        "transformer_blocks.9.ff_context.net.0.proj",
        "single_transformer_blocks.36.attn.to_v",
        "transformer_blocks.6.norm1.linear",
        "transformer_blocks.15.norm1.linear",
        "single_transformer_blocks.16.proj_out",
        "single_transformer_blocks.15.attn.to_v",
        "transformer_blocks.10.attn.add_q_proj",
        "transformer_blocks.2.norm1.linear",
        "single_transformer_blocks.29.proj_out",
        "single_transformer_blocks.28.attn.to_v",
        "single_transformer_blocks.0.proj_out",
        "single_transformer_blocks.28.proj_out",
        "single_transformer_blocks.21.norm.linear",
        "single_transformer_blocks.16.proj_mlp",
        "single_transformer_blocks.33.attn.to_q",
        "single_transformer_blocks.36.proj_mlp",
        "single_transformer_blocks.36.proj_out",
        "transformer_blocks.11.attn.add_v_proj",
        "single_transformer_blocks.26.attn.to_q",
        "transformer_blocks.8.attn.to_add_out",
        "transformer_blocks.18.attn.to_k",
        "single_transformer_blocks.31.attn.to_q",
        "single_transformer_blocks.2.attn.to_k",
        "single_transformer_blocks.10.attn.to_k",
        "transformer_blocks.4.ff_context.net.0.proj",
        "single_transformer_blocks.37.proj_mlp",
        "transformer_blocks.5.attn.add_q_proj",
        "single_transformer_blocks.27.proj_out",
        "transformer_blocks.5.ff.net.2",
        "transformer_blocks.4.attn.add_v_proj",
        "transformer_blocks.17.attn.to_out.0",
        "transformer_blocks.0.ff_context.net.0.proj",
        "transformer_blocks.11.ff.net.2",
        "transformer_blocks.9.attn.to_out.0",
        "single_transformer_blocks.32.proj_mlp",
        "transformer_blocks.3.norm1.linear",
        "single_transformer_blocks.2.attn.to_v",
        "single_transformer_blocks.12.attn.to_v",
        "single_transformer_blocks.21.proj_out",
        "transformer_blocks.5.attn.add_v_proj",
        "transformer_blocks.14.attn.to_add_out",
        "single_transformer_blocks.7.attn.to_k",
        "single_transformer_blocks.11.attn.to_v",
        "single_transformer_blocks.4.attn.to_q",
        "transformer_blocks.10.attn.to_k",
        "single_transformer_blocks.17.attn.to_k",
        "single_transformer_blocks.30.norm.linear",
        "transformer_blocks.9.attn.add_q_proj",
        "transformer_blocks.4.norm1.linear",
        "single_transformer_blocks.37.attn.to_q",
        "transformer_blocks.16.ff_context.net.0.proj",
        "single_transformer_blocks.34.attn.to_q",
        "transformer_blocks.13.attn.to_k",
        "transformer_blocks.12.norm1_context.linear",
        "single_transformer_blocks.29.attn.to_v",
        "transformer_blocks.4.attn.to_out.0",
        "single_transformer_blocks.35.proj_out",
        "transformer_blocks.8.attn.to_q",
        "single_transformer_blocks.23.norm.linear",
        "transformer_blocks.0.ff_context.net.2",
        "transformer_blocks.14.norm1_context.linear",
        "transformer_blocks.16.ff.net.0.proj",
        "transformer_blocks.10.attn.add_k_proj",
        "single_transformer_blocks.34.norm.linear",
        "transformer_blocks.17.attn.to_q",
        "single_transformer_blocks.6.norm.linear",
        "transformer_blocks.1.ff.net.0.proj",
        "single_transformer_blocks.15.proj_mlp",
        "transformer_blocks.16.ff_context.net.2",
        "single_transformer_blocks.8.attn.to_q",
        "transformer_blocks.3.attn.to_add_out",
        "single_transformer_blocks.20.proj_out",
        "single_transformer_blocks.26.attn.to_v",
        "single_transformer_blocks.32.attn.to_k",
        "single_transformer_blocks.16.attn.to_v",
        "single_transformer_blocks.17.norm.linear",
        "transformer_blocks.13.norm1_context.linear",
        "transformer_blocks.3.ff.net.2",
        "transformer_blocks.10.ff_context.net.0.proj",
        "transformer_blocks.18.attn.add_k_proj",
        "transformer_blocks.0.attn.to_v",
        "single_transformer_blocks.12.attn.to_k",
        "single_transformer_blocks.19.proj_mlp",
        "transformer_blocks.7.ff_context.net.2",
        "single_transformer_blocks.25.norm.linear",
        "transformer_blocks.0.norm1_context.linear",
        "transformer_blocks.9.attn.to_v",
        "transformer_blocks.7.ff.net.0.proj",
        "single_transformer_blocks.30.proj_out",
        "single_transformer_blocks.7.norm.linear",
        "single_transformer_blocks.31.attn.to_v",
        "transformer_blocks.11.norm1_context.linear",
        "transformer_blocks.15.ff.net.2",
        "single_transformer_blocks.28.norm.linear",
        "transformer_blocks.17.ff_context.net.2",
        "single_transformer_blocks.2.proj_out",
        "single_transformer_blocks.2.norm.linear",
        "transformer_blocks.14.attn.to_v",
        "single_transformer_blocks.19.attn.to_k",
        "single_transformer_blocks.18.attn.to_k",
        "transformer_blocks.8.attn.to_k",
        "transformer_blocks.8.norm1_context.linear",
        "transformer_blocks.11.attn.add_k_proj",
        "single_transformer_blocks.1.proj_out",
        "single_transformer_blocks.25.attn.to_k",
        "transformer_blocks.8.attn.to_v",
        "single_transformer_blocks.28.proj_mlp",
        "transformer_blocks.4.ff.net.0.proj",
        "transformer_blocks.15.attn.to_v",
        "single_transformer_blocks.24.proj_mlp",
        "transformer_blocks.5.attn.to_q",
        "transformer_blocks.10.ff_context.net.2",
        "single_transformer_blocks.9.proj_mlp",
        "single_transformer_blocks.25.attn.to_q",
        "single_transformer_blocks.16.norm.linear",
        "single_transformer_blocks.26.proj_mlp",
        "transformer_blocks.14.attn.to_k",
        "single_transformer_blocks.6.attn.to_v",
        "single_transformer_blocks.24.norm.linear",
        "transformer_blocks.2.attn.add_k_proj",
        "single_transformer_blocks.17.attn.to_v",
        "transformer_blocks.10.attn.to_v",
        "single_transformer_blocks.1.attn.to_q",
        "transformer_blocks.18.ff.net.2",
        "single_transformer_blocks.4.norm.linear",
        "transformer_blocks.9.norm1_context.linear",
        "single_transformer_blocks.8.proj_out",
        "transformer_blocks.1.norm1.linear",
        "single_transformer_blocks.18.attn.to_v",
        "single_transformer_blocks.20.proj_mlp",
        "transformer_blocks.14.norm1.linear",
        "single_transformer_blocks.30.attn.to_v",
        "transformer_blocks.16.attn.to_q",
        "transformer_blocks.17.ff_context.net.0.proj",
        "transformer_blocks.6.attn.add_k_proj",
        "transformer_blocks.15.norm1_context.linear",
        "transformer_blocks.3.attn.to_q",
        "transformer_blocks.13.ff.net.2",
        "single_transformer_blocks.26.norm.linear",
        "single_transformer_blocks.31.proj_mlp",
        "single_transformer_blocks.25.proj_out",
        "transformer_blocks.1.attn.add_k_proj",
        "single_transformer_blocks.5.proj_mlp",
        "single_transformer_blocks.29.attn.to_q",
        "single_transformer_blocks.13.proj_out",
        "transformer_blocks.0.ff.net.2",
        "transformer_blocks.2.attn.to_out.0",
        "transformer_blocks.6.attn.to_k",
        "transformer_blocks.6.attn.add_q_proj",
        "single_transformer_blocks.33.proj_mlp",
        "transformer_blocks.5.norm1_context.linear",
        "transformer_blocks.7.attn.add_k_proj",
        "single_transformer_blocks.0.norm.linear",
        "transformer_blocks.9.ff_context.net.2",
        "transformer_blocks.16.attn.to_out.0",
        "single_transformer_blocks.35.attn.to_v",
        "single_transformer_blocks.0.attn.to_k",
        "single_transformer_blocks.21.attn.to_k",
        "transformer_blocks.3.attn.add_k_proj",
        "transformer_blocks.7.norm1_context.linear",
        "transformer_blocks.18.norm1.linear",
        "single_transformer_blocks.22.proj_out",
        "single_transformer_blocks.23.proj_mlp",
        "single_transformer_blocks.11.attn.to_q",
        "single_transformer_blocks.22.norm.linear",
        "transformer_blocks.1.attn.to_q",
        "transformer_blocks.0.attn.add_k_proj",
        "transformer_blocks.13.ff_context.net.0.proj",
        "single_transformer_blocks.27.proj_mlp",
        "single_transformer_blocks.3.norm.linear",
        "transformer_blocks.8.attn.add_q_proj",
        "single_transformer_blocks.8.attn.to_k",
        "single_transformer_blocks.11.proj_out",
        "transformer_blocks.14.ff.net.0.proj",
        "transformer_blocks.17.attn.to_k",
        "transformer_blocks.6.attn.to_q",
        "single_transformer_blocks.24.proj_out",
        "transformer_blocks.11.attn.to_v",
        "transformer_blocks.2.attn.to_add_out",
        "single_transformer_blocks.8.norm.linear",
        "transformer_blocks.3.attn.add_v_proj",
        "transformer_blocks.12.attn.add_v_proj",
        "single_transformer_blocks.12.proj_mlp",
        "transformer_blocks.3.ff_context.net.2",
        "single_transformer_blocks.30.proj_mlp",
        "transformer_blocks.6.attn.add_v_proj",
        "transformer_blocks.17.attn.add_k_proj",
        "transformer_blocks.15.attn.add_k_proj",
        "transformer_blocks.11.norm1.linear",
        "transformer_blocks.1.attn.add_v_proj",
        "single_transformer_blocks.7.attn.to_v",
        "single_transformer_blocks.1.attn.to_v",
        "single_transformer_blocks.7.proj_out",
        "transformer_blocks.5.ff.net.0.proj",
        "single_transformer_blocks.21.proj_mlp",
        "single_transformer_blocks.18.attn.to_q",
        "single_transformer_blocks.7.attn.to_q",
        "transformer_blocks.12.ff_context.net.0.proj",
        "single_transformer_blocks.19.attn.to_v",
        "transformer_blocks.5.attn.to_out.0",
        "single_transformer_blocks.18.norm.linear",
        "transformer_blocks.10.norm1.linear",
        "single_transformer_blocks.11.norm.linear",
        "single_transformer_blocks.32.proj_out",
        "single_transformer_blocks.6.proj_out",
        "transformer_blocks.12.attn.to_k",
        "single_transformer_blocks.2.attn.to_q",
        "transformer_blocks.3.norm1_context.linear",
        "transformer_blocks.8.attn.to_out.0",
        "transformer_blocks.13.attn.to_out.0",
        "transformer_blocks.16.norm1.linear",
        "single_transformer_blocks.35.attn.to_q",
        "transformer_blocks.7.attn.to_out.0",
        "transformer_blocks.4.attn.to_q",
        "transformer_blocks.12.attn.add_q_proj",
        "transformer_blocks.2.attn.add_q_proj",
        "transformer_blocks.14.ff_context.net.2",
        "transformer_blocks.14.attn.to_q",
        "single_transformer_blocks.13.attn.to_q",
        "transformer_blocks.18.attn.add_v_proj",
        "transformer_blocks.2.attn.add_v_proj",
        "transformer_blocks.11.attn.add_q_proj",
        "single_transformer_blocks.13.attn.to_v",
        "single_transformer_blocks.19.norm.linear",
        "single_transformer_blocks.31.norm.linear",
        "single_transformer_blocks.25.attn.to_v",
        "transformer_blocks.18.ff_context.net.0.proj",
        "single_transformer_blocks.10.attn.to_q",
        "single_transformer_blocks.13.proj_mlp",
        "transformer_blocks.7.attn.add_v_proj",
        "single_transformer_blocks.7.proj_mlp",
        "single_transformer_blocks.3.attn.to_q",
        "single_transformer_blocks.22.attn.to_k",
        "single_transformer_blocks.21.attn.to_q",
        "transformer_blocks.18.attn.to_v",
        "transformer_blocks.13.norm1.linear",
        "single_transformer_blocks.10.norm.linear",
        "transformer_blocks.14.attn.add_k_proj",
        "transformer_blocks.16.attn.add_q_proj",
        "transformer_blocks.15.ff_context.net.2",
        "transformer_blocks.10.ff.net.0.proj",
        "single_transformer_blocks.8.attn.to_v",
        "single_transformer_blocks.29.proj_mlp",
        "transformer_blocks.1.attn.to_k",
        "transformer_blocks.9.ff.net.0.proj",
        "single_transformer_blocks.34.attn.to_k",
        "transformer_blocks.16.attn.add_k_proj",
        "transformer_blocks.16.norm1_context.linear",
        "transformer_blocks.18.ff.net.0.proj",
        "transformer_blocks.6.attn.to_out.0",
        "transformer_blocks.11.attn.to_add_out",
        "transformer_blocks.1.ff_context.net.2",
        "single_transformer_blocks.20.norm.linear",
        "transformer_blocks.8.ff.net.2",
        "single_transformer_blocks.19.proj_out",
        "transformer_blocks.4.attn.add_k_proj",
        "transformer_blocks.5.attn.to_v",
        "transformer_blocks.10.attn.to_q",
        "transformer_blocks.11.ff.net.0.proj",
        "transformer_blocks.4.attn.to_k",
        "transformer_blocks.1.norm1_context.linear",
        "single_transformer_blocks.31.proj_out",
        "single_transformer_blocks.11.attn.to_k",
        "single_transformer_blocks.23.attn.to_v",
        "single_transformer_blocks.3.proj_mlp",
        "single_transformer_blocks.4.attn.to_k",
        "transformer_blocks.15.ff.net.0.proj",
        "single_transformer_blocks.27.attn.to_k",
        "transformer_blocks.3.attn.add_q_proj",
        "transformer_blocks.3.attn.to_out.0",
        "transformer_blocks.0.attn.to_add_out",
        "transformer_blocks.9.attn.to_add_out",
        "transformer_blocks.2.ff_context.net.2",
        "transformer_blocks.0.attn.to_k",
        "transformer_blocks.11.attn.to_k",
        "transformer_blocks.6.attn.to_v",
        "transformer_blocks.3.ff_context.net.0.proj",
        "single_transformer_blocks.14.proj_out",
        "transformer_blocks.17.norm1_context.linear",
        "single_transformer_blocks.12.proj_out",
        "transformer_blocks.18.attn.to_out.0",
        "single_transformer_blocks.10.proj_mlp",
        "transformer_blocks.0.norm1.linear",
        "single_transformer_blocks.22.attn.to_q",
        "transformer_blocks.16.attn.to_v",
        "transformer_blocks.7.attn.to_k",
        "single_transformer_blocks.34.proj_mlp",
        "transformer_blocks.9.ff.net.2",
        "transformer_blocks.14.attn.add_q_proj",
        "single_transformer_blocks.20.attn.to_q",
        "transformer_blocks.1.attn.to_out.0",
        "single_transformer_blocks.9.norm.linear",
        "single_transformer_blocks.34.proj_out",
        "single_transformer_blocks.9.proj_out",
        "transformer_blocks.5.attn.to_k",
        "transformer_blocks.17.ff.net.0.proj",
        "transformer_blocks.6.norm1_context.linear",
        "single_transformer_blocks.36.attn.to_k",
        "single_transformer_blocks.35.proj_mlp",
        "single_transformer_blocks.15.attn.to_k",
        "single_transformer_blocks.18.proj_out"
    ],
    "use_dora": false,
    "lora_bias": false
}
After
{
    "r": 10,
    "lora_alpha": 10,
    "rank_pattern": {
        "transformer_blocks.0.attn.to_out.0": 10,
        "transformer_blocks.0.ff.net.0.proj": 18,
        "transformer_blocks.0.ff.net.2": 11,
        "transformer_blocks.0.norm1.linear": 12,
        "transformer_blocks.0.attn.to_add_out": 3,
        "transformer_blocks.0.ff_context.net.0.proj": 2,
        "transformer_blocks.0.norm1_context.linear": 2,
        "transformer_blocks.1.attn.to_out.0": 10,
        "transformer_blocks.1.ff.net.0.proj": 29,
        "transformer_blocks.1.ff.net.2": 9,
        "transformer_blocks.1.norm1.linear": 20,
        "transformer_blocks.1.attn.to_add_out": 3,
        "transformer_blocks.1.attn.add_q_proj": 3,
        "transformer_blocks.1.attn.add_k_proj": 3,
        "transformer_blocks.1.attn.add_v_proj": 3,
        "transformer_blocks.1.ff_context.net.0.proj": 3,
        "transformer_blocks.1.ff_context.net.2": 2,
        "transformer_blocks.1.norm1_context.linear": 2,
        "transformer_blocks.2.attn.to_out.0": 13,
        "transformer_blocks.2.ff.net.0.proj": 26,
        "transformer_blocks.2.ff.net.2": 8,
        "transformer_blocks.2.norm1.linear": 20,
        "transformer_blocks.2.attn.to_add_out": 5,
        "transformer_blocks.3.attn.to_out.0": 13,
        "transformer_blocks.3.ff.net.0.proj": 11,
        "transformer_blocks.3.ff.net.2": 5,
        "transformer_blocks.3.norm1.linear": 10,
        "transformer_blocks.3.attn.to_add_out": 5,
        "transformer_blocks.3.attn.add_q_proj": 7,
        "transformer_blocks.3.attn.add_k_proj": 7,
        "transformer_blocks.3.attn.add_v_proj": 7,
        "transformer_blocks.3.ff_context.net.0.proj": 3,
        "transformer_blocks.3.ff_context.net.2": 5,
        "transformer_blocks.3.norm1_context.linear": 5,
        "transformer_blocks.4.attn.to_out.0": 8,
        "transformer_blocks.4.ff.net.0.proj": 8,
        "transformer_blocks.4.ff.net.2": 6,
        "transformer_blocks.4.norm1.linear": 8,
        "transformer_blocks.4.attn.to_add_out": 5,
        "transformer_blocks.4.attn.add_q_proj": 6,
        "transformer_blocks.4.attn.add_k_proj": 6,
        "transformer_blocks.4.attn.add_v_proj": 6,
        "transformer_blocks.4.ff_context.net.0.proj": 3,
        "transformer_blocks.4.norm1_context.linear": 5,
        "transformer_blocks.5.attn.to_out.0": 9,
        "transformer_blocks.5.ff.net.0.proj": 10,
        "transformer_blocks.5.ff.net.2": 7,
        "transformer_blocks.5.norm1.linear": 8,
        "transformer_blocks.5.attn.add_q_proj": 3,
        "transformer_blocks.5.attn.add_k_proj": 3,
        "transformer_blocks.5.attn.add_v_proj": 3,
        "transformer_blocks.5.ff_context.net.2": 2,
        "transformer_blocks.5.norm1_context.linear": 3,
        "transformer_blocks.6.attn.to_out.0": 11,
        "transformer_blocks.6.ff.net.0.proj": 11,
        "transformer_blocks.6.ff.net.2": 5,
        "transformer_blocks.6.norm1.linear": 8,
        "transformer_blocks.6.attn.to_add_out": 5,
        "transformer_blocks.6.attn.add_q_proj": 3,
        "transformer_blocks.6.attn.add_k_proj": 3,
        "transformer_blocks.6.attn.add_v_proj": 3,
        "transformer_blocks.6.ff_context.net.0.proj": 5,
        "transformer_blocks.6.norm1_context.linear": 2,
        "transformer_blocks.7.attn.to_out.0": 12,
        "transformer_blocks.7.ff.net.0.proj": 12,
        "transformer_blocks.7.ff.net.2": 5,
        "transformer_blocks.7.norm1.linear": 8,
        "transformer_blocks.7.attn.add_q_proj": 5,
        "transformer_blocks.7.attn.add_k_proj": 5,
        "transformer_blocks.7.attn.add_v_proj": 5,
        "transformer_blocks.7.ff_context.net.0.proj": 3,
        "transformer_blocks.7.norm1_context.linear": 6,
        "transformer_blocks.8.attn.to_out.0": 13,
        "transformer_blocks.8.ff.net.0.proj": 9,
        "transformer_blocks.8.ff.net.2": 6,
        "transformer_blocks.8.norm1.linear": 7,
        "transformer_blocks.8.attn.to_add_out": 7,
        "transformer_blocks.8.ff_context.net.2": 6,
        "transformer_blocks.8.norm1_context.linear": 3,
        "transformer_blocks.9.attn.to_out.0": 7,
        "transformer_blocks.9.ff.net.0.proj": 12,
        "transformer_blocks.9.ff.net.2": 6,
        "transformer_blocks.9.norm1.linear": 8,
        "transformer_blocks.9.attn.to_add_out": 7,
        "transformer_blocks.9.attn.add_q_proj": 6,
        "transformer_blocks.9.attn.add_k_proj": 6,
        "transformer_blocks.9.attn.add_v_proj": 6,
        "transformer_blocks.9.ff_context.net.0.proj": 3,
        "transformer_blocks.9.ff_context.net.2": 5,
        "transformer_blocks.10.attn.to_out.0": 6,
        "transformer_blocks.10.ff.net.0.proj": 8,
        "transformer_blocks.10.ff.net.2": 5,
        "transformer_blocks.10.norm1.linear": 7,
        "transformer_blocks.10.attn.to_add_out": 6,
        "transformer_blocks.10.attn.add_q_proj": 7,
        "transformer_blocks.10.attn.add_k_proj": 7,
        "transformer_blocks.10.attn.add_v_proj": 7,
        "transformer_blocks.10.ff_context.net.0.proj": 3,
        "transformer_blocks.10.norm1_context.linear": 7,
        "transformer_blocks.11.attn.to_out.0": 9,
        "transformer_blocks.11.ff.net.0.proj": 10,
        "transformer_blocks.11.ff.net.2": 6,
        "transformer_blocks.11.norm1.linear": 7,
        "transformer_blocks.11.attn.to_add_out": 3,
        "transformer_blocks.11.attn.add_q_proj": 6,
        "transformer_blocks.11.attn.add_k_proj": 6,
        "transformer_blocks.11.attn.add_v_proj": 6,
        "transformer_blocks.11.ff_context.net.0.proj": 3,
        "transformer_blocks.11.ff_context.net.2": 5,
        "transformer_blocks.12.attn.to_out.0": 11,
        "transformer_blocks.12.ff.net.0.proj": 12,
        "transformer_blocks.12.ff.net.2": 6,
        "transformer_blocks.12.norm1.linear": 7,
        "transformer_blocks.12.attn.to_add_out": 5,
        "transformer_blocks.12.attn.add_q_proj": 5,
        "transformer_blocks.12.attn.add_k_proj": 5,
        "transformer_blocks.12.attn.add_v_proj": 5,
        "transformer_blocks.12.ff_context.net.0.proj": 3,
        "transformer_blocks.12.norm1_context.linear": 5,
        "transformer_blocks.13.attn.to_out.0": 8,
        "transformer_blocks.13.ff.net.0.proj": 9,
        "transformer_blocks.13.norm1.linear": 5,
        "transformer_blocks.13.attn.to_add_out": 5,
        "transformer_blocks.13.ff_context.net.0.proj": 2,
        "transformer_blocks.13.ff_context.net.2": 3,
        "transformer_blocks.13.norm1_context.linear": 6,
        "transformer_blocks.14.attn.to_out.0": 11,
        "transformer_blocks.14.ff.net.0.proj": 12,
        "transformer_blocks.14.ff.net.2": 6,
        "transformer_blocks.14.norm1.linear": 5,
        "transformer_blocks.14.attn.to_add_out": 3,
        "transformer_blocks.14.attn.add_q_proj": 5,
        "transformer_blocks.14.attn.add_k_proj": 5,
        "transformer_blocks.14.attn.add_v_proj": 5,
        "transformer_blocks.14.ff_context.net.0.proj": 3,
        "transformer_blocks.14.norm1_context.linear": 3,
        "transformer_blocks.15.attn.to_out.0": 9,
        "transformer_blocks.15.ff.net.0.proj": 12,
        "transformer_blocks.15.ff.net.2": 7,
        "transformer_blocks.15.norm1.linear": 6,
        "transformer_blocks.15.attn.add_q_proj": 3,
        "transformer_blocks.15.attn.add_k_proj": 3,
        "transformer_blocks.15.attn.add_v_proj": 3,
        "transformer_blocks.16.attn.to_out.0": 10,
        "transformer_blocks.16.ff.net.0.proj": 12,
        "transformer_blocks.16.ff.net.2": 7,
        "transformer_blocks.16.norm1.linear": 6,
        "transformer_blocks.16.attn.add_q_proj": 6,
        "transformer_blocks.16.attn.add_k_proj": 6,
        "transformer_blocks.16.attn.add_v_proj": 6,
        "transformer_blocks.16.ff_context.net.2": 5,
        "transformer_blocks.16.norm1_context.linear": 3,
        "transformer_blocks.17.attn.to_out.0": 8,
        "transformer_blocks.17.ff.net.0.proj": 9,
        "transformer_blocks.17.ff.net.2": 9,
        "transformer_blocks.17.norm1.linear": 8,
        "transformer_blocks.17.attn.to_add_out": 5,
        "transformer_blocks.17.attn.add_q_proj": 5,
        "transformer_blocks.17.attn.add_k_proj": 5,
        "transformer_blocks.17.attn.add_v_proj": 5,
        "transformer_blocks.17.ff_context.net.0.proj": 2,
        "transformer_blocks.17.norm1_context.linear": 2,
        "transformer_blocks.18.attn.to_out.0": 5,
        "transformer_blocks.18.ff.net.0.proj": 11,
        "transformer_blocks.18.ff.net.2": 8,
        "transformer_blocks.18.norm1.linear": 6,
        "transformer_blocks.18.ff_context.net.0.proj": 3,
        "transformer_blocks.18.norm1_context.linear": 7,
        "single_transformer_blocks.0.attn.to_q": 18,
        "single_transformer_blocks.0.attn.to_k": 18,
        "single_transformer_blocks.0.attn.to_v": 18,
        "single_transformer_blocks.0.proj_mlp": 18,
        "single_transformer_blocks.0.proj_out": 8,
        "single_transformer_blocks.0.norm.linear": 7,
        "single_transformer_blocks.1.attn.to_q": 22,
        "single_transformer_blocks.1.attn.to_k": 22,
        "single_transformer_blocks.1.attn.to_v": 22,
        "single_transformer_blocks.1.proj_mlp": 22,
        "single_transformer_blocks.1.proj_out": 7,
        "single_transformer_blocks.1.norm.linear": 7,
        "single_transformer_blocks.2.attn.to_q": 18,
        "single_transformer_blocks.2.attn.to_k": 18,
        "single_transformer_blocks.2.attn.to_v": 18,
        "single_transformer_blocks.2.proj_mlp": 18,
        "single_transformer_blocks.2.proj_out": 7,
        "single_transformer_blocks.2.norm.linear": 7,
        "single_transformer_blocks.3.attn.to_q": 21,
        "single_transformer_blocks.3.attn.to_k": 21,
        "single_transformer_blocks.3.attn.to_v": 21,
        "single_transformer_blocks.3.proj_mlp": 21,
        "single_transformer_blocks.3.proj_out": 7,
        "single_transformer_blocks.3.norm.linear": 8,
        "single_transformer_blocks.4.attn.to_q": 18,
        "single_transformer_blocks.4.attn.to_k": 18,
        "single_transformer_blocks.4.attn.to_v": 18,
        "single_transformer_blocks.4.proj_mlp": 18,
        "single_transformer_blocks.4.proj_out": 8,
        "single_transformer_blocks.4.norm.linear": 8,
        "single_transformer_blocks.5.attn.to_q": 20,
        "single_transformer_blocks.5.attn.to_k": 20,
        "single_transformer_blocks.5.attn.to_v": 20,
        "single_transformer_blocks.5.proj_mlp": 20,
        "single_transformer_blocks.5.proj_out": 8,
        "single_transformer_blocks.5.norm.linear": 9,
        "single_transformer_blocks.6.attn.to_q": 20,
        "single_transformer_blocks.6.attn.to_k": 20,
        "single_transformer_blocks.6.attn.to_v": 20,
        "single_transformer_blocks.6.proj_mlp": 20,
        "single_transformer_blocks.6.proj_out": 6,
        "single_transformer_blocks.6.norm.linear": 10,
        "single_transformer_blocks.7.attn.to_q": 21,
        "single_transformer_blocks.7.attn.to_k": 21,
        "single_transformer_blocks.7.attn.to_v": 21,
        "single_transformer_blocks.7.proj_mlp": 21,
        "single_transformer_blocks.7.proj_out": 8,
        "single_transformer_blocks.7.norm.linear": 10,
        "single_transformer_blocks.8.attn.to_q": 19,
        "single_transformer_blocks.8.attn.to_k": 19,
        "single_transformer_blocks.8.attn.to_v": 19,
        "single_transformer_blocks.8.proj_mlp": 19,
        "single_transformer_blocks.8.proj_out": 7,
        "single_transformer_blocks.8.norm.linear": 7,
        "single_transformer_blocks.9.attn.to_q": 22,
        "single_transformer_blocks.9.attn.to_k": 22,
        "single_transformer_blocks.9.attn.to_v": 22,
        "single_transformer_blocks.9.proj_mlp": 22,
        "single_transformer_blocks.9.proj_out": 9,
        "single_transformer_blocks.9.norm.linear": 8,
        "single_transformer_blocks.10.attn.to_q": 22,
        "single_transformer_blocks.10.attn.to_k": 22,
        "single_transformer_blocks.10.attn.to_v": 22,
        "single_transformer_blocks.10.proj_mlp": 22,
        "single_transformer_blocks.10.proj_out": 11,
        "single_transformer_blocks.10.norm.linear": 11,
        "single_transformer_blocks.11.attn.to_q": 22,
        "single_transformer_blocks.11.attn.to_k": 22,
        "single_transformer_blocks.11.attn.to_v": 22,
        "single_transformer_blocks.11.proj_mlp": 22,
        "single_transformer_blocks.11.proj_out": 10,
        "single_transformer_blocks.11.norm.linear": 10,
        "single_transformer_blocks.12.attn.to_q": 23,
        "single_transformer_blocks.12.attn.to_k": 23,
        "single_transformer_blocks.12.attn.to_v": 23,
        "single_transformer_blocks.12.proj_mlp": 23,
        "single_transformer_blocks.12.proj_out": 10,
        "single_transformer_blocks.12.norm.linear": 8,
        "single_transformer_blocks.13.attn.to_q": 22,
        "single_transformer_blocks.13.attn.to_k": 22,
        "single_transformer_blocks.13.attn.to_v": 22,
        "single_transformer_blocks.13.proj_mlp": 22,
        "single_transformer_blocks.13.proj_out": 10,
        "single_transformer_blocks.13.norm.linear": 12,
        "single_transformer_blocks.14.attn.to_q": 24,
        "single_transformer_blocks.14.attn.to_k": 24,
        "single_transformer_blocks.14.attn.to_v": 24,
        "single_transformer_blocks.14.proj_mlp": 24,
        "single_transformer_blocks.14.proj_out": 11,
        "single_transformer_blocks.14.norm.linear": 11,
        "single_transformer_blocks.15.attn.to_q": 25,
        "single_transformer_blocks.15.attn.to_k": 25,
        "single_transformer_blocks.15.attn.to_v": 25,
        "single_transformer_blocks.15.proj_mlp": 25,
        "single_transformer_blocks.15.proj_out": 10,
        "single_transformer_blocks.15.norm.linear": 11,
        "single_transformer_blocks.16.attn.to_q": 26,
        "single_transformer_blocks.16.attn.to_k": 26,
        "single_transformer_blocks.16.attn.to_v": 26,
        "single_transformer_blocks.16.proj_mlp": 26,
        "single_transformer_blocks.16.proj_out": 10,
        "single_transformer_blocks.16.norm.linear": 9,
        "single_transformer_blocks.17.attn.to_q": 28,
        "single_transformer_blocks.17.attn.to_k": 28,
        "single_transformer_blocks.17.attn.to_v": 28,
        "single_transformer_blocks.17.proj_mlp": 28,
        "single_transformer_blocks.17.proj_out": 10,
        "single_transformer_blocks.17.norm.linear": 9,
        "single_transformer_blocks.18.attn.to_q": 28,
        "single_transformer_blocks.18.attn.to_k": 28,
        "single_transformer_blocks.18.attn.to_v": 28,
        "single_transformer_blocks.18.proj_mlp": 28,
        "single_transformer_blocks.18.proj_out": 8,
        "single_transformer_blocks.18.norm.linear": 11,
        "single_transformer_blocks.19.attn.to_q": 27,
        "single_transformer_blocks.19.attn.to_k": 27,
        "single_transformer_blocks.19.attn.to_v": 27,
        "single_transformer_blocks.19.proj_mlp": 27,
        "single_transformer_blocks.19.proj_out": 10,
        "single_transformer_blocks.19.norm.linear": 10,
        "single_transformer_blocks.20.attn.to_q": 29,
        "single_transformer_blocks.20.attn.to_k": 29,
        "single_transformer_blocks.20.attn.to_v": 29,
        "single_transformer_blocks.20.proj_mlp": 29,
        "single_transformer_blocks.20.proj_out": 11,
        "single_transformer_blocks.20.norm.linear": 12,
        "single_transformer_blocks.21.attn.to_q": 26,
        "single_transformer_blocks.21.attn.to_k": 26,
        "single_transformer_blocks.21.attn.to_v": 26,
        "single_transformer_blocks.21.proj_mlp": 26,
        "single_transformer_blocks.21.proj_out": 11,
        "single_transformer_blocks.21.norm.linear": 11,
        "single_transformer_blocks.22.attn.to_q": 26,
        "single_transformer_blocks.22.attn.to_k": 26,
        "single_transformer_blocks.22.attn.to_v": 26,
        "single_transformer_blocks.22.proj_mlp": 26,
        "single_transformer_blocks.22.proj_out": 11,
        "single_transformer_blocks.22.norm.linear": 10,
        "single_transformer_blocks.23.attn.to_q": 30,
        "single_transformer_blocks.23.attn.to_k": 30,
        "single_transformer_blocks.23.attn.to_v": 30,
        "single_transformer_blocks.23.proj_mlp": 30,
        "single_transformer_blocks.23.proj_out": 11,
        "single_transformer_blocks.23.norm.linear": 11,
        "single_transformer_blocks.24.attn.to_q": 28,
        "single_transformer_blocks.24.attn.to_k": 28,
        "single_transformer_blocks.24.attn.to_v": 28,
        "single_transformer_blocks.24.proj_mlp": 28,
        "single_transformer_blocks.24.proj_out": 12,
        "single_transformer_blocks.24.norm.linear": 11,
        "single_transformer_blocks.25.attn.to_q": 28,
        "single_transformer_blocks.25.attn.to_k": 28,
        "single_transformer_blocks.25.attn.to_v": 28,
        "single_transformer_blocks.25.proj_mlp": 28,
        "single_transformer_blocks.25.proj_out": 14,
        "single_transformer_blocks.25.norm.linear": 9,
        "single_transformer_blocks.26.attn.to_q": 28,
        "single_transformer_blocks.26.attn.to_k": 28,
        "single_transformer_blocks.26.attn.to_v": 28,
        "single_transformer_blocks.26.proj_mlp": 28,
        "single_transformer_blocks.26.proj_out": 13,
        "single_transformer_blocks.26.norm.linear": 9,
        "single_transformer_blocks.27.attn.to_q": 29,
        "single_transformer_blocks.27.attn.to_k": 29,
        "single_transformer_blocks.27.attn.to_v": 29,
        "single_transformer_blocks.27.proj_mlp": 29,
        "single_transformer_blocks.27.proj_out": 14,
        "single_transformer_blocks.27.norm.linear": 8,
        "single_transformer_blocks.28.attn.to_q": 29,
        "single_transformer_blocks.28.attn.to_k": 29,
        "single_transformer_blocks.28.attn.to_v": 29,
        "single_transformer_blocks.28.proj_mlp": 29,
        "single_transformer_blocks.28.proj_out": 14,
        "single_transformer_blocks.28.norm.linear": 12,
        "single_transformer_blocks.29.attn.to_q": 30,
        "single_transformer_blocks.29.attn.to_k": 30,
        "single_transformer_blocks.29.attn.to_v": 30,
        "single_transformer_blocks.29.proj_mlp": 30,
        "single_transformer_blocks.29.proj_out": 14,
        "single_transformer_blocks.29.norm.linear": 10,
        "single_transformer_blocks.30.attn.to_q": 32,
        "single_transformer_blocks.30.attn.to_k": 32,
        "single_transformer_blocks.30.attn.to_v": 32,
        "single_transformer_blocks.30.proj_mlp": 32,
        "single_transformer_blocks.30.proj_out": 15,
        "single_transformer_blocks.30.norm.linear": 9,
        "single_transformer_blocks.31.attn.to_q": 30,
        "single_transformer_blocks.31.attn.to_k": 30,
        "single_transformer_blocks.31.attn.to_v": 30,
        "single_transformer_blocks.31.proj_mlp": 30,
        "single_transformer_blocks.31.proj_out": 16,
        "single_transformer_blocks.31.norm.linear": 11,
        "single_transformer_blocks.32.attn.to_q": 30,
        "single_transformer_blocks.32.attn.to_k": 30,
        "single_transformer_blocks.32.attn.to_v": 30,
        "single_transformer_blocks.32.proj_mlp": 30,
        "single_transformer_blocks.32.proj_out": 15,
        "single_transformer_blocks.32.norm.linear": 9,
        "single_transformer_blocks.33.attn.to_q": 32,
        "single_transformer_blocks.33.attn.to_k": 32,
        "single_transformer_blocks.33.attn.to_v": 32,
        "single_transformer_blocks.33.proj_mlp": 32,
        "single_transformer_blocks.33.proj_out": 16,
        "single_transformer_blocks.33.norm.linear": 8,
        "single_transformer_blocks.34.attn.to_q": 30,
        "single_transformer_blocks.34.attn.to_k": 30,
        "single_transformer_blocks.34.attn.to_v": 30,
        "single_transformer_blocks.34.proj_mlp": 30,
        "single_transformer_blocks.34.proj_out": 16,
        "single_transformer_blocks.34.norm.linear": 7,
        "single_transformer_blocks.35.attn.to_q": 31,
        "single_transformer_blocks.35.attn.to_k": 31,
        "single_transformer_blocks.35.attn.to_v": 31,
        "single_transformer_blocks.35.proj_mlp": 31,
        "single_transformer_blocks.35.proj_out": 14,
        "single_transformer_blocks.35.norm.linear": 8,
        "single_transformer_blocks.36.attn.to_q": 30,
        "single_transformer_blocks.36.attn.to_k": 30,
        "single_transformer_blocks.36.attn.to_v": 30,
        "single_transformer_blocks.36.proj_mlp": 30,
        "single_transformer_blocks.36.proj_out": 12,
        "single_transformer_blocks.36.norm.linear": 9,
        "single_transformer_blocks.37.attn.to_q": 17,
        "single_transformer_blocks.37.attn.to_k": 17,
        "single_transformer_blocks.37.attn.to_v": 17,
        "single_transformer_blocks.37.proj_mlp": 17,
        "single_transformer_blocks.37.proj_out": 7,
        "single_transformer_blocks.37.norm.linear": 9,
        "transformer_blocks.8.attn.add_v_proj": 4,
        "transformer_blocks.18.attn.to_add_out": 4,
        "transformer_blocks.6.ff_context.net.2": 4,
        "transformer_blocks.8.ff_context.net.0.proj": 4,
        "transformer_blocks.13.attn.add_k_proj": 4,
        "transformer_blocks.8.attn.add_k_proj": 4,
        "transformer_blocks.5.ff_context.net.0.proj": 4,
        "transformer_blocks.12.ff_context.net.2": 4,
        "transformer_blocks.2.ff_context.net.0.proj": 4,
        "transformer_blocks.2.norm1_context.linear": 4,
        "transformer_blocks.13.attn.add_q_proj": 4,
        "transformer_blocks.15.ff_context.net.0.proj": 4,
        "transformer_blocks.18.ff_context.net.2": 4,
        "transformer_blocks.5.attn.to_add_out": 4,
        "transformer_blocks.7.attn.to_add_out": 4,
        "transformer_blocks.13.attn.add_v_proj": 4,
        "transformer_blocks.0.attn.add_v_proj": 4,
        "transformer_blocks.16.attn.to_add_out": 4,
        "transformer_blocks.4.ff_context.net.2": 4,
        "transformer_blocks.15.attn.to_add_out": 4,
        "transformer_blocks.18.attn.add_q_proj": 4,
        "transformer_blocks.0.attn.add_q_proj": 4,
        "transformer_blocks.16.ff_context.net.0.proj": 4,
        "transformer_blocks.0.ff_context.net.2": 4,
        "transformer_blocks.18.attn.add_k_proj": 4,
        "transformer_blocks.7.ff_context.net.2": 4,
        "transformer_blocks.11.norm1_context.linear": 4,
        "transformer_blocks.17.ff_context.net.2": 4,
        "transformer_blocks.10.ff_context.net.2": 4,
        "transformer_blocks.2.attn.add_k_proj": 4,
        "transformer_blocks.9.norm1_context.linear": 4,
        "transformer_blocks.15.norm1_context.linear": 4,
        "transformer_blocks.13.ff.net.2": 4,
        "transformer_blocks.0.attn.add_k_proj": 4,
        "transformer_blocks.8.attn.add_q_proj": 4,
        "transformer_blocks.2.attn.add_q_proj": 4,
        "transformer_blocks.14.ff_context.net.2": 4,
        "transformer_blocks.18.attn.add_v_proj": 4,
        "transformer_blocks.2.attn.add_v_proj": 4,
        "transformer_blocks.15.ff_context.net.2": 4,
        "transformer_blocks.2.ff_context.net.2": 4
    },
    "alpha_pattern": {
        "transformer_blocks.0.attn.to_out.0": 10,
        "transformer_blocks.0.ff.net.0.proj": 18,
        "transformer_blocks.0.ff.net.2": 11,
        "transformer_blocks.0.norm1.linear": 12,
        "transformer_blocks.0.attn.to_add_out": 3,
        "transformer_blocks.0.ff_context.net.0.proj": 2,
        "transformer_blocks.0.norm1_context.linear": 2,
        "transformer_blocks.1.attn.to_out.0": 10,
        "transformer_blocks.1.ff.net.0.proj": 29,
        "transformer_blocks.1.ff.net.2": 9,
        "transformer_blocks.1.norm1.linear": 20,
        "transformer_blocks.1.attn.to_add_out": 3,
        "transformer_blocks.1.attn.add_q_proj": 3,
        "transformer_blocks.1.attn.add_k_proj": 3,
        "transformer_blocks.1.attn.add_v_proj": 3,
        "transformer_blocks.1.ff_context.net.0.proj": 3,
        "transformer_blocks.1.ff_context.net.2": 2,
        "transformer_blocks.1.norm1_context.linear": 2,
        "transformer_blocks.2.attn.to_out.0": 13,
        "transformer_blocks.2.ff.net.0.proj": 26,
        "transformer_blocks.2.ff.net.2": 8,
        "transformer_blocks.2.norm1.linear": 20,
        "transformer_blocks.2.attn.to_add_out": 5,
        "transformer_blocks.3.attn.to_out.0": 13,
        "transformer_blocks.3.ff.net.0.proj": 11,
        "transformer_blocks.3.ff.net.2": 5,
        "transformer_blocks.3.norm1.linear": 10,
        "transformer_blocks.3.attn.to_add_out": 5,
        "transformer_blocks.3.attn.add_q_proj": 7,
        "transformer_blocks.3.attn.add_k_proj": 7,
        "transformer_blocks.3.attn.add_v_proj": 7,
        "transformer_blocks.3.ff_context.net.0.proj": 3,
        "transformer_blocks.3.ff_context.net.2": 5,
        "transformer_blocks.3.norm1_context.linear": 5,
        "transformer_blocks.4.attn.to_out.0": 8,
        "transformer_blocks.4.ff.net.0.proj": 8,
        "transformer_blocks.4.ff.net.2": 6,
        "transformer_blocks.4.norm1.linear": 8,
        "transformer_blocks.4.attn.to_add_out": 5,
        "transformer_blocks.4.attn.add_q_proj": 6,
        "transformer_blocks.4.attn.add_k_proj": 6,
        "transformer_blocks.4.attn.add_v_proj": 6,
        "transformer_blocks.4.ff_context.net.0.proj": 3,
        "transformer_blocks.4.norm1_context.linear": 5,
        "transformer_blocks.5.attn.to_out.0": 9,
        "transformer_blocks.5.ff.net.0.proj": 10,
        "transformer_blocks.5.ff.net.2": 7,
        "transformer_blocks.5.norm1.linear": 8,
        "transformer_blocks.5.attn.add_q_proj": 3,
        "transformer_blocks.5.attn.add_k_proj": 3,
        "transformer_blocks.5.attn.add_v_proj": 3,
        "transformer_blocks.5.ff_context.net.2": 2,
        "transformer_blocks.5.norm1_context.linear": 3,
        "transformer_blocks.6.attn.to_out.0": 11,
        "transformer_blocks.6.ff.net.0.proj": 11,
        "transformer_blocks.6.ff.net.2": 5,
        "transformer_blocks.6.norm1.linear": 8,
        "transformer_blocks.6.attn.to_add_out": 5,
        "transformer_blocks.6.attn.add_q_proj": 3,
        "transformer_blocks.6.attn.add_k_proj": 3,
        "transformer_blocks.6.attn.add_v_proj": 3,
        "transformer_blocks.6.ff_context.net.0.proj": 5,
        "transformer_blocks.6.norm1_context.linear": 2,
        "transformer_blocks.7.attn.to_out.0": 12,
        "transformer_blocks.7.ff.net.0.proj": 12,
        "transformer_blocks.7.ff.net.2": 5,
        "transformer_blocks.7.norm1.linear": 8,
        "transformer_blocks.7.attn.add_q_proj": 5,
        "transformer_blocks.7.attn.add_k_proj": 5,
        "transformer_blocks.7.attn.add_v_proj": 5,
        "transformer_blocks.7.ff_context.net.0.proj": 3,
        "transformer_blocks.7.norm1_context.linear": 6,
        "transformer_blocks.8.attn.to_out.0": 13,
        "transformer_blocks.8.ff.net.0.proj": 9,
        "transformer_blocks.8.ff.net.2": 6,
        "transformer_blocks.8.norm1.linear": 7,
        "transformer_blocks.8.attn.to_add_out": 7,
        "transformer_blocks.8.ff_context.net.2": 6,
        "transformer_blocks.8.norm1_context.linear": 3,
        "transformer_blocks.9.attn.to_out.0": 7,
        "transformer_blocks.9.ff.net.0.proj": 12,
        "transformer_blocks.9.ff.net.2": 6,
        "transformer_blocks.9.norm1.linear": 8,
        "transformer_blocks.9.attn.to_add_out": 7,
        "transformer_blocks.9.attn.add_q_proj": 6,
        "transformer_blocks.9.attn.add_k_proj": 6,
        "transformer_blocks.9.attn.add_v_proj": 6,
        "transformer_blocks.9.ff_context.net.0.proj": 3,
        "transformer_blocks.9.ff_context.net.2": 5,
        "transformer_blocks.10.attn.to_out.0": 6,
        "transformer_blocks.10.ff.net.0.proj": 8,
        "transformer_blocks.10.ff.net.2": 5,
        "transformer_blocks.10.norm1.linear": 7,
        "transformer_blocks.10.attn.to_add_out": 6,
        "transformer_blocks.10.attn.add_q_proj": 7,
        "transformer_blocks.10.attn.add_k_proj": 7,
        "transformer_blocks.10.attn.add_v_proj": 7,
        "transformer_blocks.10.ff_context.net.0.proj": 3,
        "transformer_blocks.10.norm1_context.linear": 7,
        "transformer_blocks.11.attn.to_out.0": 9,
        "transformer_blocks.11.ff.net.0.proj": 10,
        "transformer_blocks.11.ff.net.2": 6,
        "transformer_blocks.11.norm1.linear": 7,
        "transformer_blocks.11.attn.to_add_out": 3,
        "transformer_blocks.11.attn.add_q_proj": 6,
        "transformer_blocks.11.attn.add_k_proj": 6,
        "transformer_blocks.11.attn.add_v_proj": 6,
        "transformer_blocks.11.ff_context.net.0.proj": 3,
        "transformer_blocks.11.ff_context.net.2": 5,
        "transformer_blocks.12.attn.to_out.0": 11,
        "transformer_blocks.12.ff.net.0.proj": 12,
        "transformer_blocks.12.ff.net.2": 6,
        "transformer_blocks.12.norm1.linear": 7,
        "transformer_blocks.12.attn.to_add_out": 5,
        "transformer_blocks.12.attn.add_q_proj": 5,
        "transformer_blocks.12.attn.add_k_proj": 5,
        "transformer_blocks.12.attn.add_v_proj": 5,
        "transformer_blocks.12.ff_context.net.0.proj": 3,
        "transformer_blocks.12.norm1_context.linear": 5,
        "transformer_blocks.13.attn.to_out.0": 8,
        "transformer_blocks.13.ff.net.0.proj": 9,
        "transformer_blocks.13.norm1.linear": 5,
        "transformer_blocks.13.attn.to_add_out": 5,
        "transformer_blocks.13.ff_context.net.0.proj": 2,
        "transformer_blocks.13.ff_context.net.2": 3,
        "transformer_blocks.13.norm1_context.linear": 6,
        "transformer_blocks.14.attn.to_out.0": 11,
        "transformer_blocks.14.ff.net.0.proj": 12,
        "transformer_blocks.14.ff.net.2": 6,
        "transformer_blocks.14.norm1.linear": 5,
        "transformer_blocks.14.attn.to_add_out": 3,
        "transformer_blocks.14.attn.add_q_proj": 5,
        "transformer_blocks.14.attn.add_k_proj": 5,
        "transformer_blocks.14.attn.add_v_proj": 5,
        "transformer_blocks.14.ff_context.net.0.proj": 3,
        "transformer_blocks.14.norm1_context.linear": 3,
        "transformer_blocks.15.attn.to_out.0": 9,
        "transformer_blocks.15.ff.net.0.proj": 12,
        "transformer_blocks.15.ff.net.2": 7,
        "transformer_blocks.15.norm1.linear": 6,
        "transformer_blocks.15.attn.add_q_proj": 3,
        "transformer_blocks.15.attn.add_k_proj": 3,
        "transformer_blocks.15.attn.add_v_proj": 3,
        "transformer_blocks.16.attn.to_out.0": 10,
        "transformer_blocks.16.ff.net.0.proj": 12,
        "transformer_blocks.16.ff.net.2": 7,
        "transformer_blocks.16.norm1.linear": 6,
        "transformer_blocks.16.attn.add_q_proj": 6,
        "transformer_blocks.16.attn.add_k_proj": 6,
        "transformer_blocks.16.attn.add_v_proj": 6,
        "transformer_blocks.16.ff_context.net.2": 5,
        "transformer_blocks.16.norm1_context.linear": 3,
        "transformer_blocks.17.attn.to_out.0": 8,
        "transformer_blocks.17.ff.net.0.proj": 9,
        "transformer_blocks.17.ff.net.2": 9,
        "transformer_blocks.17.norm1.linear": 8,
        "transformer_blocks.17.attn.to_add_out": 5,
        "transformer_blocks.17.attn.add_q_proj": 5,
        "transformer_blocks.17.attn.add_k_proj": 5,
        "transformer_blocks.17.attn.add_v_proj": 5,
        "transformer_blocks.17.ff_context.net.0.proj": 2,
        "transformer_blocks.17.norm1_context.linear": 2,
        "transformer_blocks.18.attn.to_out.0": 5,
        "transformer_blocks.18.ff.net.0.proj": 11,
        "transformer_blocks.18.ff.net.2": 8,
        "transformer_blocks.18.norm1.linear": 6,
        "transformer_blocks.18.ff_context.net.0.proj": 3,
        "transformer_blocks.18.norm1_context.linear": 7,
        "single_transformer_blocks.0.attn.to_q": 18,
        "single_transformer_blocks.0.attn.to_k": 18,
        "single_transformer_blocks.0.attn.to_v": 18,
        "single_transformer_blocks.0.proj_mlp": 18,
        "single_transformer_blocks.0.proj_out": 8,
        "single_transformer_blocks.0.norm.linear": 7,
        "single_transformer_blocks.1.attn.to_q": 22,
        "single_transformer_blocks.1.attn.to_k": 22,
        "single_transformer_blocks.1.attn.to_v": 22,
        "single_transformer_blocks.1.proj_mlp": 22,
        "single_transformer_blocks.1.proj_out": 7,
        "single_transformer_blocks.1.norm.linear": 7,
        "single_transformer_blocks.2.attn.to_q": 18,
        "single_transformer_blocks.2.attn.to_k": 18,
        "single_transformer_blocks.2.attn.to_v": 18,
        "single_transformer_blocks.2.proj_mlp": 18,
        "single_transformer_blocks.2.proj_out": 7,
        "single_transformer_blocks.2.norm.linear": 7,
        "single_transformer_blocks.3.attn.to_q": 21,
        "single_transformer_blocks.3.attn.to_k": 21,
        "single_transformer_blocks.3.attn.to_v": 21,
        "single_transformer_blocks.3.proj_mlp": 21,
        "single_transformer_blocks.3.proj_out": 7,
        "single_transformer_blocks.3.norm.linear": 8,
        "single_transformer_blocks.4.attn.to_q": 18,
        "single_transformer_blocks.4.attn.to_k": 18,
        "single_transformer_blocks.4.attn.to_v": 18,
        "single_transformer_blocks.4.proj_mlp": 18,
        "single_transformer_blocks.4.proj_out": 8,
        "single_transformer_blocks.4.norm.linear": 8,
        "single_transformer_blocks.5.attn.to_q": 20,
        "single_transformer_blocks.5.attn.to_k": 20,
        "single_transformer_blocks.5.attn.to_v": 20,
        "single_transformer_blocks.5.proj_mlp": 20,
        "single_transformer_blocks.5.proj_out": 8,
        "single_transformer_blocks.5.norm.linear": 9,
        "single_transformer_blocks.6.attn.to_q": 20,
        "single_transformer_blocks.6.attn.to_k": 20,
        "single_transformer_blocks.6.attn.to_v": 20,
        "single_transformer_blocks.6.proj_mlp": 20,
        "single_transformer_blocks.6.proj_out": 6,
        "single_transformer_blocks.6.norm.linear": 10,
        "single_transformer_blocks.7.attn.to_q": 21,
        "single_transformer_blocks.7.attn.to_k": 21,
        "single_transformer_blocks.7.attn.to_v": 21,
        "single_transformer_blocks.7.proj_mlp": 21,
        "single_transformer_blocks.7.proj_out": 8,
        "single_transformer_blocks.7.norm.linear": 10,
        "single_transformer_blocks.8.attn.to_q": 19,
        "single_transformer_blocks.8.attn.to_k": 19,
        "single_transformer_blocks.8.attn.to_v": 19,
        "single_transformer_blocks.8.proj_mlp": 19,
        "single_transformer_blocks.8.proj_out": 7,
        "single_transformer_blocks.8.norm.linear": 7,
        "single_transformer_blocks.9.attn.to_q": 22,
        "single_transformer_blocks.9.attn.to_k": 22,
        "single_transformer_blocks.9.attn.to_v": 22,
        "single_transformer_blocks.9.proj_mlp": 22,
        "single_transformer_blocks.9.proj_out": 9,
        "single_transformer_blocks.9.norm.linear": 8,
        "single_transformer_blocks.10.attn.to_q": 22,
        "single_transformer_blocks.10.attn.to_k": 22,
        "single_transformer_blocks.10.attn.to_v": 22,
        "single_transformer_blocks.10.proj_mlp": 22,
        "single_transformer_blocks.10.proj_out": 11,
        "single_transformer_blocks.10.norm.linear": 11,
        "single_transformer_blocks.11.attn.to_q": 22,
        "single_transformer_blocks.11.attn.to_k": 22,
        "single_transformer_blocks.11.attn.to_v": 22,
        "single_transformer_blocks.11.proj_mlp": 22,
        "single_transformer_blocks.11.proj_out": 10,
        "single_transformer_blocks.11.norm.linear": 10,
        "single_transformer_blocks.12.attn.to_q": 23,
        "single_transformer_blocks.12.attn.to_k": 23,
        "single_transformer_blocks.12.attn.to_v": 23,
        "single_transformer_blocks.12.proj_mlp": 23,
        "single_transformer_blocks.12.proj_out": 10,
        "single_transformer_blocks.12.norm.linear": 8,
        "single_transformer_blocks.13.attn.to_q": 22,
        "single_transformer_blocks.13.attn.to_k": 22,
        "single_transformer_blocks.13.attn.to_v": 22,
        "single_transformer_blocks.13.proj_mlp": 22,
        "single_transformer_blocks.13.proj_out": 10,
        "single_transformer_blocks.13.norm.linear": 12,
        "single_transformer_blocks.14.attn.to_q": 24,
        "single_transformer_blocks.14.attn.to_k": 24,
        "single_transformer_blocks.14.attn.to_v": 24,
        "single_transformer_blocks.14.proj_mlp": 24,
        "single_transformer_blocks.14.proj_out": 11,
        "single_transformer_blocks.14.norm.linear": 11,
        "single_transformer_blocks.15.attn.to_q": 25,
        "single_transformer_blocks.15.attn.to_k": 25,
        "single_transformer_blocks.15.attn.to_v": 25,
        "single_transformer_blocks.15.proj_mlp": 25,
        "single_transformer_blocks.15.proj_out": 10,
        "single_transformer_blocks.15.norm.linear": 11,
        "single_transformer_blocks.16.attn.to_q": 26,
        "single_transformer_blocks.16.attn.to_k": 26,
        "single_transformer_blocks.16.attn.to_v": 26,
        "single_transformer_blocks.16.proj_mlp": 26,
        "single_transformer_blocks.16.proj_out": 10,
        "single_transformer_blocks.16.norm.linear": 9,
        "single_transformer_blocks.17.attn.to_q": 28,
        "single_transformer_blocks.17.attn.to_k": 28,
        "single_transformer_blocks.17.attn.to_v": 28,
        "single_transformer_blocks.17.proj_mlp": 28,
        "single_transformer_blocks.17.proj_out": 10,
        "single_transformer_blocks.17.norm.linear": 9,
        "single_transformer_blocks.18.attn.to_q": 28,
        "single_transformer_blocks.18.attn.to_k": 28,
        "single_transformer_blocks.18.attn.to_v": 28,
        "single_transformer_blocks.18.proj_mlp": 28,
        "single_transformer_blocks.18.proj_out": 8,
        "single_transformer_blocks.18.norm.linear": 11,
        "single_transformer_blocks.19.attn.to_q": 27,
        "single_transformer_blocks.19.attn.to_k": 27,
        "single_transformer_blocks.19.attn.to_v": 27,
        "single_transformer_blocks.19.proj_mlp": 27,
        "single_transformer_blocks.19.proj_out": 10,
        "single_transformer_blocks.19.norm.linear": 10,
        "single_transformer_blocks.20.attn.to_q": 29,
        "single_transformer_blocks.20.attn.to_k": 29,
        "single_transformer_blocks.20.attn.to_v": 29,
        "single_transformer_blocks.20.proj_mlp": 29,
        "single_transformer_blocks.20.proj_out": 11,
        "single_transformer_blocks.20.norm.linear": 12,
        "single_transformer_blocks.21.attn.to_q": 26,
        "single_transformer_blocks.21.attn.to_k": 26,
        "single_transformer_blocks.21.attn.to_v": 26,
        "single_transformer_blocks.21.proj_mlp": 26,
        "single_transformer_blocks.21.proj_out": 11,
        "single_transformer_blocks.21.norm.linear": 11,
        "single_transformer_blocks.22.attn.to_q": 26,
        "single_transformer_blocks.22.attn.to_k": 26,
        "single_transformer_blocks.22.attn.to_v": 26,
        "single_transformer_blocks.22.proj_mlp": 26,
        "single_transformer_blocks.22.proj_out": 11,
        "single_transformer_blocks.22.norm.linear": 10,
        "single_transformer_blocks.23.attn.to_q": 30,
        "single_transformer_blocks.23.attn.to_k": 30,
        "single_transformer_blocks.23.attn.to_v": 30,
        "single_transformer_blocks.23.proj_mlp": 30,
        "single_transformer_blocks.23.proj_out": 11,
        "single_transformer_blocks.23.norm.linear": 11,
        "single_transformer_blocks.24.attn.to_q": 28,
        "single_transformer_blocks.24.attn.to_k": 28,
        "single_transformer_blocks.24.attn.to_v": 28,
        "single_transformer_blocks.24.proj_mlp": 28,
        "single_transformer_blocks.24.proj_out": 12,
        "single_transformer_blocks.24.norm.linear": 11,
        "single_transformer_blocks.25.attn.to_q": 28,
        "single_transformer_blocks.25.attn.to_k": 28,
        "single_transformer_blocks.25.attn.to_v": 28,
        "single_transformer_blocks.25.proj_mlp": 28,
        "single_transformer_blocks.25.proj_out": 14,
        "single_transformer_blocks.25.norm.linear": 9,
        "single_transformer_blocks.26.attn.to_q": 28,
        "single_transformer_blocks.26.attn.to_k": 28,
        "single_transformer_blocks.26.attn.to_v": 28,
        "single_transformer_blocks.26.proj_mlp": 28,
        "single_transformer_blocks.26.proj_out": 13,
        "single_transformer_blocks.26.norm.linear": 9,
        "single_transformer_blocks.27.attn.to_q": 29,
        "single_transformer_blocks.27.attn.to_k": 29,
        "single_transformer_blocks.27.attn.to_v": 29,
        "single_transformer_blocks.27.proj_mlp": 29,
        "single_transformer_blocks.27.proj_out": 14,
        "single_transformer_blocks.27.norm.linear": 8,
        "single_transformer_blocks.28.attn.to_q": 29,
        "single_transformer_blocks.28.attn.to_k": 29,
        "single_transformer_blocks.28.attn.to_v": 29,
        "single_transformer_blocks.28.proj_mlp": 29,
        "single_transformer_blocks.28.proj_out": 14,
        "single_transformer_blocks.28.norm.linear": 12,
        "single_transformer_blocks.29.attn.to_q": 30,
        "single_transformer_blocks.29.attn.to_k": 30,
        "single_transformer_blocks.29.attn.to_v": 30,
        "single_transformer_blocks.29.proj_mlp": 30,
        "single_transformer_blocks.29.proj_out": 14,
        "single_transformer_blocks.29.norm.linear": 10,
        "single_transformer_blocks.30.attn.to_q": 32,
        "single_transformer_blocks.30.attn.to_k": 32,
        "single_transformer_blocks.30.attn.to_v": 32,
        "single_transformer_blocks.30.proj_mlp": 32,
        "single_transformer_blocks.30.proj_out": 15,
        "single_transformer_blocks.30.norm.linear": 9,
        "single_transformer_blocks.31.attn.to_q": 30,
        "single_transformer_blocks.31.attn.to_k": 30,
        "single_transformer_blocks.31.attn.to_v": 30,
        "single_transformer_blocks.31.proj_mlp": 30,
        "single_transformer_blocks.31.proj_out": 16,
        "single_transformer_blocks.31.norm.linear": 11,
        "single_transformer_blocks.32.attn.to_q": 30,
        "single_transformer_blocks.32.attn.to_k": 30,
        "single_transformer_blocks.32.attn.to_v": 30,
        "single_transformer_blocks.32.proj_mlp": 30,
        "single_transformer_blocks.32.proj_out": 15,
        "single_transformer_blocks.32.norm.linear": 9,
        "single_transformer_blocks.33.attn.to_q": 32,
        "single_transformer_blocks.33.attn.to_k": 32,
        "single_transformer_blocks.33.attn.to_v": 32,
        "single_transformer_blocks.33.proj_mlp": 32,
        "single_transformer_blocks.33.proj_out": 16,
        "single_transformer_blocks.33.norm.linear": 8,
        "single_transformer_blocks.34.attn.to_q": 30,
        "single_transformer_blocks.34.attn.to_k": 30,
        "single_transformer_blocks.34.attn.to_v": 30,
        "single_transformer_blocks.34.proj_mlp": 30,
        "single_transformer_blocks.34.proj_out": 16,
        "single_transformer_blocks.34.norm.linear": 7,
        "single_transformer_blocks.35.attn.to_q": 31,
        "single_transformer_blocks.35.attn.to_k": 31,
        "single_transformer_blocks.35.attn.to_v": 31,
        "single_transformer_blocks.35.proj_mlp": 31,
        "single_transformer_blocks.35.proj_out": 14,
        "single_transformer_blocks.35.norm.linear": 8,
        "single_transformer_blocks.36.attn.to_q": 30,
        "single_transformer_blocks.36.attn.to_k": 30,
        "single_transformer_blocks.36.attn.to_v": 30,
        "single_transformer_blocks.36.proj_mlp": 30,
        "single_transformer_blocks.36.proj_out": 12,
        "single_transformer_blocks.36.norm.linear": 9,
        "single_transformer_blocks.37.attn.to_q": 17,
        "single_transformer_blocks.37.attn.to_k": 17,
        "single_transformer_blocks.37.attn.to_v": 17,
        "single_transformer_blocks.37.proj_mlp": 17,
        "single_transformer_blocks.37.proj_out": 7,
        "single_transformer_blocks.37.norm.linear": 9,
        "transformer_blocks.8.attn.add_v_proj": 4,
        "transformer_blocks.18.attn.to_add_out": 4,
        "transformer_blocks.6.ff_context.net.2": 4,
        "transformer_blocks.8.ff_context.net.0.proj": 4,
        "transformer_blocks.13.attn.add_k_proj": 4,
        "transformer_blocks.8.attn.add_k_proj": 4,
        "transformer_blocks.5.ff_context.net.0.proj": 4,
        "transformer_blocks.12.ff_context.net.2": 4,
        "transformer_blocks.2.ff_context.net.0.proj": 4,
        "transformer_blocks.2.norm1_context.linear": 4,
        "transformer_blocks.13.attn.add_q_proj": 4,
        "transformer_blocks.15.ff_context.net.0.proj": 4,
        "transformer_blocks.18.ff_context.net.2": 4,
        "transformer_blocks.5.attn.to_add_out": 4,
        "transformer_blocks.7.attn.to_add_out": 4,
        "transformer_blocks.13.attn.add_v_proj": 4,
        "transformer_blocks.0.attn.add_v_proj": 4,
        "transformer_blocks.16.attn.to_add_out": 4,
        "transformer_blocks.4.ff_context.net.2": 4,
        "transformer_blocks.15.attn.to_add_out": 4,
        "transformer_blocks.18.attn.add_q_proj": 4,
        "transformer_blocks.0.attn.add_q_proj": 4,
        "transformer_blocks.16.ff_context.net.0.proj": 4,
        "transformer_blocks.0.ff_context.net.2": 4,
        "transformer_blocks.18.attn.add_k_proj": 4,
        "transformer_blocks.7.ff_context.net.2": 4,
        "transformer_blocks.11.norm1_context.linear": 4,
        "transformer_blocks.17.ff_context.net.2": 4,
        "transformer_blocks.10.ff_context.net.2": 4,
        "transformer_blocks.2.attn.add_k_proj": 4,
        "transformer_blocks.9.norm1_context.linear": 4,
        "transformer_blocks.15.norm1_context.linear": 4,
        "transformer_blocks.13.ff.net.2": 4,
        "transformer_blocks.0.attn.add_k_proj": 4,
        "transformer_blocks.8.attn.add_q_proj": 4,
        "transformer_blocks.2.attn.add_q_proj": 4,
        "transformer_blocks.14.ff_context.net.2": 4,
        "transformer_blocks.18.attn.add_v_proj": 4,
        "transformer_blocks.2.attn.add_v_proj": 4,
        "transformer_blocks.15.ff_context.net.2": 4,
        "transformer_blocks.2.ff_context.net.2": 4
    },
    "target_modules": [
        "transformer_blocks.11.ff_context.net.2",
        "single_transformer_blocks.23.attn.to_k",
        "transformer_blocks.8.attn.add_v_proj",
        "single_transformer_blocks.5.proj_out",
        "transformer_blocks.18.attn.to_add_out",
        "single_transformer_blocks.33.attn.to_v",
        "transformer_blocks.17.norm1.linear",
        "single_transformer_blocks.1.proj_mlp",
        "transformer_blocks.4.attn.to_v",
        "transformer_blocks.7.attn.to_v",
        "transformer_blocks.13.attn.to_add_out",
        "transformer_blocks.10.ff.net.2",
        "single_transformer_blocks.0.attn.to_v",
        "single_transformer_blocks.22.attn.to_v",
        "transformer_blocks.6.ff_context.net.2",
        "transformer_blocks.15.attn.to_q",
        "single_transformer_blocks.25.proj_mlp",
        "single_transformer_blocks.17.attn.to_q",
        "single_transformer_blocks.26.proj_out",
        "single_transformer_blocks.30.attn.to_k",
        "single_transformer_blocks.32.attn.to_v",
        "transformer_blocks.1.ff.net.2",
        "single_transformer_blocks.10.proj_out",
        "single_transformer_blocks.14.norm.linear",
        "transformer_blocks.15.attn.to_out.0",
        "transformer_blocks.17.ff.net.2",
        "single_transformer_blocks.34.attn.to_v",
        "single_transformer_blocks.4.proj_mlp",
        "transformer_blocks.7.attn.to_q",
        "transformer_blocks.0.ff.net.0.proj",
        "transformer_blocks.14.attn.add_v_proj",
        "transformer_blocks.8.ff_context.net.0.proj",
        "single_transformer_blocks.28.attn.to_q",
        "transformer_blocks.13.attn.add_k_proj",
        "single_transformer_blocks.2.proj_mlp",
        "single_transformer_blocks.21.attn.to_v",
        "single_transformer_blocks.9.attn.to_k",
        "single_transformer_blocks.9.attn.to_v",
        "transformer_blocks.17.attn.to_v",
        "single_transformer_blocks.31.attn.to_k",
        "transformer_blocks.10.norm1_context.linear",
        "transformer_blocks.17.attn.add_q_proj",
        "transformer_blocks.6.ff.net.2",
        "transformer_blocks.12.ff.net.0.proj",
        "single_transformer_blocks.1.attn.to_k",
        "transformer_blocks.1.attn.to_v",
        "single_transformer_blocks.23.attn.to_q",
        "transformer_blocks.13.ff.net.0.proj",
        "single_transformer_blocks.17.proj_mlp",
        "transformer_blocks.8.attn.add_k_proj",
        "transformer_blocks.5.ff_context.net.0.proj",
        "single_transformer_blocks.5.norm.linear",
        "transformer_blocks.15.attn.to_k",
        "single_transformer_blocks.20.attn.to_v",
        "single_transformer_blocks.36.attn.to_q",
        "transformer_blocks.9.norm1.linear",
        "single_transformer_blocks.5.attn.to_q",
        "single_transformer_blocks.10.attn.to_v",
        "single_transformer_blocks.30.attn.to_q",
        "transformer_blocks.12.attn.to_add_out",
        "transformer_blocks.12.ff_context.net.2",
        "transformer_blocks.11.ff_context.net.0.proj",
        "transformer_blocks.2.ff_context.net.0.proj",
        "single_transformer_blocks.6.attn.to_q",
        "single_transformer_blocks.3.attn.to_k",
        "single_transformer_blocks.13.attn.to_k",
        "single_transformer_blocks.16.attn.to_q",
        "single_transformer_blocks.27.attn.to_q",
        "transformer_blocks.7.ff_context.net.0.proj",
        "single_transformer_blocks.23.proj_out",
        "transformer_blocks.12.attn.add_k_proj",
        "single_transformer_blocks.15.attn.to_q",
        "single_transformer_blocks.37.proj_out",
        "transformer_blocks.3.attn.to_v",
        "transformer_blocks.17.attn.add_v_proj",
        "transformer_blocks.2.norm1_context.linear",
        "single_transformer_blocks.32.norm.linear",
        "transformer_blocks.12.ff.net.2",
        "single_transformer_blocks.20.attn.to_k",
        "transformer_blocks.2.ff.net.2",
        "single_transformer_blocks.3.proj_out",
        "transformer_blocks.18.attn.to_q",
        "single_transformer_blocks.14.attn.to_q",
        "single_transformer_blocks.11.proj_mlp",
        "single_transformer_blocks.14.attn.to_v",
        "transformer_blocks.13.attn.add_q_proj",
        "transformer_blocks.15.ff_context.net.0.proj",
        "single_transformer_blocks.8.proj_mlp",
        "single_transformer_blocks.16.attn.to_k",
        "single_transformer_blocks.27.norm.linear",
        "transformer_blocks.8.ff.net.0.proj",
        "single_transformer_blocks.33.norm.linear",
        "transformer_blocks.10.attn.to_add_out",
        "single_transformer_blocks.33.proj_out",
        "transformer_blocks.16.attn.add_v_proj",
        "transformer_blocks.12.attn.to_out.0",
        "transformer_blocks.15.attn.add_q_proj",
        "transformer_blocks.2.attn.to_v",
        "single_transformer_blocks.17.proj_out",
        "transformer_blocks.5.attn.add_k_proj",
        "transformer_blocks.12.attn.to_q",
        "single_transformer_blocks.0.attn.to_q",
        "transformer_blocks.1.attn.add_q_proj",
        "transformer_blocks.14.attn.to_out.0",
        "single_transformer_blocks.33.attn.to_k",
        "transformer_blocks.3.ff.net.0.proj",
        "single_transformer_blocks.18.proj_mlp",
        "single_transformer_blocks.26.attn.to_k",
        "transformer_blocks.4.attn.add_q_proj",
        "transformer_blocks.10.attn.to_out.0",
        "transformer_blocks.13.ff_context.net.2",
        "transformer_blocks.6.ff.net.0.proj",
        "transformer_blocks.2.attn.to_q",
        "transformer_blocks.17.attn.to_add_out",
        "single_transformer_blocks.24.attn.to_q",
        "single_transformer_blocks.32.attn.to_q",
        "transformer_blocks.16.ff.net.2",
        "single_transformer_blocks.12.norm.linear",
        "transformer_blocks.4.attn.to_add_out",
        "single_transformer_blocks.28.attn.to_k",
        "transformer_blocks.18.ff_context.net.2",
        "transformer_blocks.9.attn.to_q",
        "transformer_blocks.8.norm1.linear",
        "transformer_blocks.16.attn.to_k",
        "transformer_blocks.13.attn.to_q",
        "transformer_blocks.5.attn.to_add_out",
        "transformer_blocks.7.attn.to_add_out",
        "transformer_blocks.9.attn.add_v_proj",
        "transformer_blocks.1.attn.to_add_out",
        "transformer_blocks.1.ff_context.net.0.proj",
        "transformer_blocks.2.ff.net.0.proj",
        "transformer_blocks.7.ff.net.2",
        "transformer_blocks.13.attn.add_v_proj",
        "single_transformer_blocks.35.attn.to_k",
        "single_transformer_blocks.29.norm.linear",
        "transformer_blocks.11.attn.to_out.0",
        "transformer_blocks.8.ff_context.net.2",
        "transformer_blocks.12.norm1.linear",
        "transformer_blocks.7.norm1.linear",
        "single_transformer_blocks.9.attn.to_q",
        "transformer_blocks.9.attn.add_k_proj",
        "single_transformer_blocks.14.attn.to_k",
        "single_transformer_blocks.3.attn.to_v",
        "single_transformer_blocks.37.norm.linear",
        "transformer_blocks.13.attn.to_v",
        "single_transformer_blocks.0.proj_mlp",
        "transformer_blocks.0.attn.to_q",
        "transformer_blocks.6.attn.to_add_out",
        "transformer_blocks.12.attn.to_v",
        "single_transformer_blocks.4.proj_out",
        "single_transformer_blocks.24.attn.to_v",
        "single_transformer_blocks.4.attn.to_v",
        "transformer_blocks.3.attn.to_k",
        "transformer_blocks.5.ff_context.net.2",
        "single_transformer_blocks.15.proj_out",
        "transformer_blocks.14.ff.net.2",
        "transformer_blocks.0.attn.add_v_proj",
        "single_transformer_blocks.15.norm.linear",
        "single_transformer_blocks.19.attn.to_q",
        "single_transformer_blocks.35.norm.linear",
        "transformer_blocks.11.attn.to_q",
        "single_transformer_blocks.6.proj_mlp",
        "transformer_blocks.16.attn.to_add_out",
        "single_transformer_blocks.14.proj_mlp",
        "transformer_blocks.10.attn.add_v_proj",
        "transformer_blocks.4.ff_context.net.2",
        "transformer_blocks.15.attn.to_add_out",
        "single_transformer_blocks.27.attn.to_v",
        "transformer_blocks.2.attn.to_k",
        "transformer_blocks.7.attn.add_q_proj",
        "single_transformer_blocks.1.norm.linear",
        "single_transformer_blocks.24.attn.to_k",
        "transformer_blocks.18.attn.add_q_proj",
        "transformer_blocks.14.ff_context.net.0.proj",
        "single_transformer_blocks.6.attn.to_k",
        "single_transformer_blocks.37.attn.to_v",
        "transformer_blocks.0.attn.add_q_proj",
        "transformer_blocks.0.attn.to_out.0",
        "single_transformer_blocks.5.attn.to_v",
        "single_transformer_blocks.5.attn.to_k",
        "transformer_blocks.18.norm1_context.linear",
        "transformer_blocks.4.ff.net.2",
        "single_transformer_blocks.29.attn.to_k",
        "transformer_blocks.6.ff_context.net.0.proj",
        "transformer_blocks.9.attn.to_k",
        "single_transformer_blocks.12.attn.to_q",
        "single_transformer_blocks.37.attn.to_k",
        "single_transformer_blocks.22.proj_mlp",
        "transformer_blocks.15.attn.add_v_proj",
        "transformer_blocks.4.norm1_context.linear",
        "single_transformer_blocks.13.norm.linear",
        "single_transformer_blocks.36.norm.linear",
        "transformer_blocks.5.norm1.linear",
        "transformer_blocks.9.ff_context.net.0.proj",
        "single_transformer_blocks.36.attn.to_v",
        "transformer_blocks.6.norm1.linear",
        "transformer_blocks.15.norm1.linear",
        "single_transformer_blocks.16.proj_out",
        "single_transformer_blocks.15.attn.to_v",
        "transformer_blocks.10.attn.add_q_proj",
        "transformer_blocks.2.norm1.linear",
        "single_transformer_blocks.29.proj_out",
        "single_transformer_blocks.28.attn.to_v",
        "single_transformer_blocks.0.proj_out",
        "single_transformer_blocks.28.proj_out",
        "single_transformer_blocks.21.norm.linear",
        "single_transformer_blocks.16.proj_mlp",
        "single_transformer_blocks.33.attn.to_q",
        "single_transformer_blocks.36.proj_mlp",
        "single_transformer_blocks.36.proj_out",
        "transformer_blocks.11.attn.add_v_proj",
        "single_transformer_blocks.26.attn.to_q",
        "transformer_blocks.8.attn.to_add_out",
        "transformer_blocks.18.attn.to_k",
        "single_transformer_blocks.31.attn.to_q",
        "single_transformer_blocks.2.attn.to_k",
        "single_transformer_blocks.10.attn.to_k",
        "transformer_blocks.4.ff_context.net.0.proj",
        "single_transformer_blocks.37.proj_mlp",
        "transformer_blocks.5.attn.add_q_proj",
        "single_transformer_blocks.27.proj_out",
        "transformer_blocks.5.ff.net.2",
        "transformer_blocks.4.attn.add_v_proj",
        "transformer_blocks.17.attn.to_out.0",
        "transformer_blocks.0.ff_context.net.0.proj",
        "transformer_blocks.11.ff.net.2",
        "transformer_blocks.9.attn.to_out.0",
        "single_transformer_blocks.32.proj_mlp",
        "transformer_blocks.3.norm1.linear",
        "single_transformer_blocks.2.attn.to_v",
        "single_transformer_blocks.12.attn.to_v",
        "single_transformer_blocks.21.proj_out",
        "transformer_blocks.5.attn.add_v_proj",
        "transformer_blocks.14.attn.to_add_out",
        "single_transformer_blocks.7.attn.to_k",
        "single_transformer_blocks.11.attn.to_v",
        "single_transformer_blocks.4.attn.to_q",
        "transformer_blocks.10.attn.to_k",
        "single_transformer_blocks.17.attn.to_k",
        "single_transformer_blocks.30.norm.linear",
        "transformer_blocks.9.attn.add_q_proj",
        "transformer_blocks.4.norm1.linear",
        "single_transformer_blocks.37.attn.to_q",
        "transformer_blocks.16.ff_context.net.0.proj",
        "single_transformer_blocks.34.attn.to_q",
        "transformer_blocks.13.attn.to_k",
        "transformer_blocks.12.norm1_context.linear",
        "single_transformer_blocks.29.attn.to_v",
        "transformer_blocks.4.attn.to_out.0",
        "single_transformer_blocks.35.proj_out",
        "transformer_blocks.8.attn.to_q",
        "single_transformer_blocks.23.norm.linear",
        "transformer_blocks.0.ff_context.net.2",
        "transformer_blocks.14.norm1_context.linear",
        "transformer_blocks.16.ff.net.0.proj",
        "transformer_blocks.10.attn.add_k_proj",
        "single_transformer_blocks.34.norm.linear",
        "transformer_blocks.17.attn.to_q",
        "single_transformer_blocks.6.norm.linear",
        "transformer_blocks.1.ff.net.0.proj",
        "single_transformer_blocks.15.proj_mlp",
        "transformer_blocks.16.ff_context.net.2",
        "single_transformer_blocks.8.attn.to_q",
        "transformer_blocks.3.attn.to_add_out",
        "single_transformer_blocks.20.proj_out",
        "single_transformer_blocks.26.attn.to_v",
        "single_transformer_blocks.32.attn.to_k",
        "single_transformer_blocks.16.attn.to_v",
        "single_transformer_blocks.17.norm.linear",
        "transformer_blocks.13.norm1_context.linear",
        "transformer_blocks.3.ff.net.2",
        "transformer_blocks.10.ff_context.net.0.proj",
        "transformer_blocks.18.attn.add_k_proj",
        "transformer_blocks.0.attn.to_v",
        "single_transformer_blocks.12.attn.to_k",
        "single_transformer_blocks.19.proj_mlp",
        "transformer_blocks.7.ff_context.net.2",
        "single_transformer_blocks.25.norm.linear",
        "transformer_blocks.0.norm1_context.linear",
        "transformer_blocks.9.attn.to_v",
        "transformer_blocks.7.ff.net.0.proj",
        "single_transformer_blocks.30.proj_out",
        "single_transformer_blocks.7.norm.linear",
        "single_transformer_blocks.31.attn.to_v",
        "transformer_blocks.11.norm1_context.linear",
        "transformer_blocks.15.ff.net.2",
        "single_transformer_blocks.28.norm.linear",
        "transformer_blocks.17.ff_context.net.2",
        "single_transformer_blocks.2.proj_out",
        "single_transformer_blocks.2.norm.linear",
        "transformer_blocks.14.attn.to_v",
        "single_transformer_blocks.19.attn.to_k",
        "single_transformer_blocks.18.attn.to_k",
        "transformer_blocks.8.attn.to_k",
        "transformer_blocks.8.norm1_context.linear",
        "transformer_blocks.11.attn.add_k_proj",
        "single_transformer_blocks.1.proj_out",
        "single_transformer_blocks.25.attn.to_k",
        "transformer_blocks.8.attn.to_v",
        "single_transformer_blocks.28.proj_mlp",
        "transformer_blocks.4.ff.net.0.proj",
        "transformer_blocks.15.attn.to_v",
        "single_transformer_blocks.24.proj_mlp",
        "transformer_blocks.5.attn.to_q",
        "transformer_blocks.10.ff_context.net.2",
        "single_transformer_blocks.9.proj_mlp",
        "single_transformer_blocks.25.attn.to_q",
        "single_transformer_blocks.16.norm.linear",
        "single_transformer_blocks.26.proj_mlp",
        "transformer_blocks.14.attn.to_k",
        "single_transformer_blocks.6.attn.to_v",
        "single_transformer_blocks.24.norm.linear",
        "transformer_blocks.2.attn.add_k_proj",
        "single_transformer_blocks.17.attn.to_v",
        "transformer_blocks.10.attn.to_v",
        "single_transformer_blocks.1.attn.to_q",
        "transformer_blocks.18.ff.net.2",
        "single_transformer_blocks.4.norm.linear",
        "transformer_blocks.9.norm1_context.linear",
        "single_transformer_blocks.8.proj_out",
        "transformer_blocks.1.norm1.linear",
        "single_transformer_blocks.18.attn.to_v",
        "single_transformer_blocks.20.proj_mlp",
        "transformer_blocks.14.norm1.linear",
        "single_transformer_blocks.30.attn.to_v",
        "transformer_blocks.16.attn.to_q",
        "transformer_blocks.17.ff_context.net.0.proj",
        "transformer_blocks.6.attn.add_k_proj",
        "transformer_blocks.15.norm1_context.linear",
        "transformer_blocks.3.attn.to_q",
        "transformer_blocks.13.ff.net.2",
        "single_transformer_blocks.26.norm.linear",
        "single_transformer_blocks.31.proj_mlp",
        "single_transformer_blocks.25.proj_out",
        "transformer_blocks.1.attn.add_k_proj",
        "single_transformer_blocks.5.proj_mlp",
        "single_transformer_blocks.29.attn.to_q",
        "single_transformer_blocks.13.proj_out",
        "transformer_blocks.0.ff.net.2",
        "transformer_blocks.2.attn.to_out.0",
        "transformer_blocks.6.attn.to_k",
        "transformer_blocks.6.attn.add_q_proj",
        "single_transformer_blocks.33.proj_mlp",
        "transformer_blocks.5.norm1_context.linear",
        "transformer_blocks.7.attn.add_k_proj",
        "single_transformer_blocks.0.norm.linear",
        "transformer_blocks.9.ff_context.net.2",
        "transformer_blocks.16.attn.to_out.0",
        "single_transformer_blocks.35.attn.to_v",
        "single_transformer_blocks.0.attn.to_k",
        "single_transformer_blocks.21.attn.to_k",
        "transformer_blocks.3.attn.add_k_proj",
        "transformer_blocks.7.norm1_context.linear",
        "transformer_blocks.18.norm1.linear",
        "single_transformer_blocks.22.proj_out",
        "single_transformer_blocks.23.proj_mlp",
        "single_transformer_blocks.11.attn.to_q",
        "single_transformer_blocks.22.norm.linear",
        "transformer_blocks.1.attn.to_q",
        "transformer_blocks.0.attn.add_k_proj",
        "transformer_blocks.13.ff_context.net.0.proj",
        "single_transformer_blocks.27.proj_mlp",
        "single_transformer_blocks.3.norm.linear",
        "transformer_blocks.8.attn.add_q_proj",
        "single_transformer_blocks.8.attn.to_k",
        "single_transformer_blocks.11.proj_out",
        "transformer_blocks.14.ff.net.0.proj",
        "transformer_blocks.17.attn.to_k",
        "transformer_blocks.6.attn.to_q",
        "single_transformer_blocks.24.proj_out",
        "transformer_blocks.11.attn.to_v",
        "transformer_blocks.2.attn.to_add_out",
        "single_transformer_blocks.8.norm.linear",
        "transformer_blocks.3.attn.add_v_proj",
        "transformer_blocks.12.attn.add_v_proj",
        "single_transformer_blocks.12.proj_mlp",
        "transformer_blocks.3.ff_context.net.2",
        "single_transformer_blocks.30.proj_mlp",
        "transformer_blocks.6.attn.add_v_proj",
        "transformer_blocks.17.attn.add_k_proj",
        "transformer_blocks.15.attn.add_k_proj",
        "transformer_blocks.11.norm1.linear",
        "transformer_blocks.1.attn.add_v_proj",
        "single_transformer_blocks.7.attn.to_v",
        "single_transformer_blocks.1.attn.to_v",
        "single_transformer_blocks.7.proj_out",
        "transformer_blocks.5.ff.net.0.proj",
        "single_transformer_blocks.21.proj_mlp",
        "single_transformer_blocks.18.attn.to_q",
        "single_transformer_blocks.7.attn.to_q",
        "transformer_blocks.12.ff_context.net.0.proj",
        "single_transformer_blocks.19.attn.to_v",
        "transformer_blocks.5.attn.to_out.0",
        "single_transformer_blocks.18.norm.linear",
        "transformer_blocks.10.norm1.linear",
        "single_transformer_blocks.11.norm.linear",
        "single_transformer_blocks.32.proj_out",
        "single_transformer_blocks.6.proj_out",
        "transformer_blocks.12.attn.to_k",
        "single_transformer_blocks.2.attn.to_q",
        "transformer_blocks.3.norm1_context.linear",
        "transformer_blocks.8.attn.to_out.0",
        "transformer_blocks.13.attn.to_out.0",
        "transformer_blocks.16.norm1.linear",
        "single_transformer_blocks.35.attn.to_q",
        "transformer_blocks.7.attn.to_out.0",
        "transformer_blocks.4.attn.to_q",
        "transformer_blocks.12.attn.add_q_proj",
        "transformer_blocks.2.attn.add_q_proj",
        "transformer_blocks.14.ff_context.net.2",
        "transformer_blocks.14.attn.to_q",
        "single_transformer_blocks.13.attn.to_q",
        "transformer_blocks.18.attn.add_v_proj",
        "transformer_blocks.2.attn.add_v_proj",
        "transformer_blocks.11.attn.add_q_proj",
        "single_transformer_blocks.13.attn.to_v",
        "single_transformer_blocks.19.norm.linear",
        "single_transformer_blocks.31.norm.linear",
        "single_transformer_blocks.25.attn.to_v",
        "transformer_blocks.18.ff_context.net.0.proj",
        "single_transformer_blocks.10.attn.to_q",
        "single_transformer_blocks.13.proj_mlp",
        "transformer_blocks.7.attn.add_v_proj",
        "single_transformer_blocks.7.proj_mlp",
        "single_transformer_blocks.3.attn.to_q",
        "single_transformer_blocks.22.attn.to_k",
        "single_transformer_blocks.21.attn.to_q",
        "transformer_blocks.18.attn.to_v",
        "transformer_blocks.13.norm1.linear",
        "single_transformer_blocks.10.norm.linear",
        "transformer_blocks.14.attn.add_k_proj",
        "transformer_blocks.16.attn.add_q_proj",
        "transformer_blocks.15.ff_context.net.2",
        "transformer_blocks.10.ff.net.0.proj",
        "single_transformer_blocks.8.attn.to_v",
        "single_transformer_blocks.29.proj_mlp",
        "transformer_blocks.1.attn.to_k",
        "transformer_blocks.9.ff.net.0.proj",
        "single_transformer_blocks.34.attn.to_k",
        "transformer_blocks.16.attn.add_k_proj",
        "transformer_blocks.16.norm1_context.linear",
        "transformer_blocks.18.ff.net.0.proj",
        "transformer_blocks.6.attn.to_out.0",
        "transformer_blocks.11.attn.to_add_out",
        "transformer_blocks.1.ff_context.net.2",
        "single_transformer_blocks.20.norm.linear",
        "transformer_blocks.8.ff.net.2",
        "single_transformer_blocks.19.proj_out",
        "transformer_blocks.4.attn.add_k_proj",
        "transformer_blocks.5.attn.to_v",
        "transformer_blocks.10.attn.to_q",
        "transformer_blocks.11.ff.net.0.proj",
        "transformer_blocks.4.attn.to_k",
        "transformer_blocks.1.norm1_context.linear",
        "single_transformer_blocks.31.proj_out",
        "single_transformer_blocks.11.attn.to_k",
        "single_transformer_blocks.23.attn.to_v",
        "single_transformer_blocks.3.proj_mlp",
        "single_transformer_blocks.4.attn.to_k",
        "transformer_blocks.15.ff.net.0.proj",
        "single_transformer_blocks.27.attn.to_k",
        "transformer_blocks.3.attn.add_q_proj",
        "transformer_blocks.3.attn.to_out.0",
        "transformer_blocks.0.attn.to_add_out",
        "transformer_blocks.9.attn.to_add_out",
        "transformer_blocks.2.ff_context.net.2",
        "transformer_blocks.0.attn.to_k",
        "transformer_blocks.11.attn.to_k",
        "transformer_blocks.6.attn.to_v",
        "transformer_blocks.3.ff_context.net.0.proj",
        "single_transformer_blocks.14.proj_out",
        "transformer_blocks.17.norm1_context.linear",
        "single_transformer_blocks.12.proj_out",
        "transformer_blocks.18.attn.to_out.0",
        "single_transformer_blocks.10.proj_mlp",
        "transformer_blocks.0.norm1.linear",
        "single_transformer_blocks.22.attn.to_q",
        "transformer_blocks.16.attn.to_v",
        "transformer_blocks.7.attn.to_k",
        "single_transformer_blocks.34.proj_mlp",
        "transformer_blocks.9.ff.net.2",
        "transformer_blocks.14.attn.add_q_proj",
        "single_transformer_blocks.20.attn.to_q",
        "transformer_blocks.1.attn.to_out.0",
        "single_transformer_blocks.9.norm.linear",
        "single_transformer_blocks.34.proj_out",
        "single_transformer_blocks.9.proj_out",
        "transformer_blocks.5.attn.to_k",
        "transformer_blocks.17.ff.net.0.proj",
        "transformer_blocks.6.norm1_context.linear",
        "single_transformer_blocks.36.attn.to_k",
        "single_transformer_blocks.35.proj_mlp",
        "single_transformer_blocks.15.attn.to_k",
        "single_transformer_blocks.18.proj_out"
    ],
    "use_dora": false,
    "lora_bias": false
}

We see the QKV modules for transformer_blocks being purged from the rank_pattern. This is what creates the mess.

@sayakpaul sayakpaul marked this pull request as ready for review March 18, 2025 09:01
@sayakpaul sayakpaul requested review from DN6 and hlky March 18, 2025 09:02
@sayakpaul sayakpaul changed the title [LoRA] support more comyui loras for Flux. [LoRA] support more comyui loras for Flux 🚨 Mar 18, 2025
Copy link
Contributor

@hlky hlky left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @sayakpaul

@sayakpaul
Copy link
Member Author

@BenjaminBossan I asked for a review from you for peft related bit.s

Copy link
Member

@BenjaminBossan BenjaminBossan left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Generally LGTM, thanks. I added some comments but they're not blockers.


has_norm_diff = any("norm" in k and "diff" in k for k in state_dict)
if has_norm_diff:
zero_status_diff = state_dict_all_zero(state_dict, "diff")
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not familiar with those other state dict formats, just wanted to ask whether it would be safer to use dots in the filter keys, e.g. .diff. instead of diff to prevent accidental matches.

@lordsoffallen
Copy link

Hey @sayakpaul,

I have tested this on quite some LoRas and it works on most of them. Great news 🎉 I did find some that still does not work. I am sharing them here for visibility.

https://civitai.com/models/832858?modelVersionId=951509
https://civitai.com/models/830230?modelVersionId=954701
https://civitai.com/models/810000?modelVersionId=1264088

@sayakpaul
Copy link
Member Author

@lordsoffallen just pushed some changes that should make it possible to load the first two LoRAs you mentioned. Will work on fixing the third one later. Since the PR is already getting delayed a bit I would prefer doing that in a separate PR. Hope that is fine.

@DN6 @BenjaminBossan I have made the change requested by @BenjaminBossan in #10985 (comment). I have also run the integration tests and they pass.

LMK if you would like to do another pass before we merge this.

@sayakpaul sayakpaul merged commit 6bfacf0 into main Apr 9, 2025
31 of 32 checks passed
@sayakpaul sayakpaul deleted the support-comyui-flux-loras branch April 9, 2025 03:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

6 participants