Open
Description
Name and Version
Image: https://github.com/ggml-org/llama.cpp/pkgs/container/llama.cpp/421179669?tag=server-cuda-b5452
Operating systems
Linux
GGML backends
CUDA
Hardware
L40S GPU (aws g6e.xlarge instance)
Models
Problem description & steps to reproduce
Command: docker run -p 8080:8080 -it --gpus all ghcr.io/ggml-org/llama.cpp:server-cuda --port 8080 --host 0.0.0.0 --parallel 16 --ctx-size 49152 --cont-batching --slot-prompt-similarity 0.3 --n-gpu-layers 100000000 --flash-attn --no-warmup --jinja --lora-init-without-apply --lora-scaled ... -m ...
After the high-load requests (50 CCU), the server response empty content even the completion_tokens is ok
This is the difference in log:
The error version:
llama_kv_cache_unified_iswa: creating non-SWA KV cache, size = 49152 cells
llama_kv_cache_unified: CUDA0 KV buffer size = 3840.00 MiB
llama_kv_cache_unified: size = 3840.00 MiB ( 49152 cells, 10 layers), K (f16): 1920.00 MiB, V (f16): 1920.00 MiB
llama_kv_cache_unified_iswa: creating SWA KV cache, size = 18432 cells
llama_kv_cache_unified: CUDA0 KV buffer size = 7488.00 MiB
llama_kv_cache_unified: size = 7488.00 MiB ( 18432 cells, 52 layers), K (f16): 3744.00 MiB, V (f16): 3744.00 MiB
llama_context: CUDA0 compute buffer size = 130.62 MiB
llama_context: CUDA_Host compute buffer size = 35.63 MiB
llama_context: graph nodes = 2489
llama_context: graph splits = 2
common_init_from_params: KV cache shifting is not supported for this context, disabling KV cache shifting
The previous version (work well):
llama_kv_cache_unified: kv_size = 49152, type_k = 'f16', type_v = 'f16', n_layer = 62, can_shift = 1, padding = 256
llama_kv_cache_unified: CUDA0 KV buffer size = 23808.00 MiB
llama_kv_cache_unified: KV self size = 23808.00 MiB, K (f16): 11904.00 MiB, V (f16): 11904.00 MiB
llama_context: CUDA0 compute buffer size = 522.50 MiB
llama_context: CUDA_Host compute buffer size = 202.51 MiB
llama_context: graph nodes = 2365
llama_context: graph splits = 2
Example response error:
{
"id": "chatcmpl-lck9Lm6T24oN61ZuwgXtfzWxYBLhKzkO",
"created": 1747909068,
"model": "google/gemma-3-27b-it",
"object": "chat.completion",
"system_fingerprint": "b5428-f0adb80b",
"choices": [
{
"finish_reason": "length",
"index": 0,
"message": {
"content": "l",
"role": "assistant",
"tool_calls": null,
"function_call": null,
"refusal": null
}
}
],
"usage": {
"completion_tokens": 45,
"prompt_tokens": 630,
"total_tokens": 675,
"completion_tokens_details": null,
"prompt_tokens_details": null
},
"service_tier": null,
"timings": {
"prompt_n": 630,
"prompt_ms": 430.044,
"prompt_per_token_ms": 0.6826095238095238,
"prompt_per_second": 1464.9663755336665,
"predicted_n": 45,
"predicted_ms": 1818.511,
"predicted_per_token_ms": 40.41135555555555,
"predicted_per_second": 24.74551982363593
}
}
First Bad Commit
This image is still working fine: https://github.com/ggml-org/llama.cpp/pkgs/container/llama.cpp/419241486?tag=server-cuda-b5428
Relevant log output
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
Device 0: NVIDIA L40S, compute capability 8.9, VMM: yes
load_backend: loaded CUDA backend from /app/libggml-cuda.so
load_backend: loaded CPU backend from /app/libggml-cpu-haswell.so
warn: LLAMA_ARG_HOST environment variable is set, but will be overwritten by command line argument --host
build: 5439 (33983057) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu
system info: n_threads = 2, n_threads_batch = 2, total_threads = 4
system_info: n_threads = 2 (n_threads_batch = 2) / 4 | CUDA : ARCHS = 500,610,700,750,800,860,890 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |
main: binding port with default address family
main: HTTP server is listening, hostname: 0.0.0.0, port: 8080, http threads: 18
main: loading model
srv load_model: loading model '/models/gemma-3-27b-it-Q4_K_M.gguf'
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA L40S) - 45036 MiB free
llama_model_loader: loaded meta data with 40 key-value pairs and 808 tensors from /models/gemma-3-27b-it-Q4_K_M.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv 0: general.architecture str = gemma3
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.name str = Gemma 3 27b It
llama_model_loader: - kv 3: general.finetune str = it
llama_model_loader: - kv 4: general.basename str = gemma-3
llama_model_loader: - kv 5: general.size_label str = 27B
llama_model_loader: - kv 6: general.license str = gemma
llama_model_loader: - kv 7: general.base_model.count u32 = 1
llama_model_loader: - kv 8: general.base_model.0.name str = Gemma 3 27b Pt
llama_model_loader: - kv 9: general.base_model.0.organization str = Google
llama_model_loader: - kv 10: general.base_model.0.repo_url str = https://huggingface.co/google/gemma-3...
llama_model_loader: - kv 11: general.tags arr[str,1] = ["image-text-to-text"]
llama_model_loader: - kv 12: gemma3.context_length u32 = 131072
llama_model_loader: - kv 13: gemma3.embedding_length u32 = 5376
llama_model_loader: - kv 14: gemma3.block_count u32 = 62
llama_model_loader: - kv 15: gemma3.feed_forward_length u32 = 21504
llama_model_loader: - kv 16: gemma3.attention.head_count u32 = 32
llama_model_loader: - kv 17: gemma3.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 18: gemma3.attention.key_length u32 = 128
llama_model_loader: - kv 19: gemma3.attention.value_length u32 = 128
llama_model_loader: - kv 20: gemma3.rope.freq_base f32 = 1000000.000000
llama_model_loader: - kv 21: gemma3.attention.sliding_window u32 = 1024
llama_model_loader: - kv 22: gemma3.attention.head_count_kv u32 = 16
llama_model_loader: - kv 23: gemma3.rope.scaling.type str = linear
llama_model_loader: - kv 24: gemma3.rope.scaling.factor f32 = 8.000000
llama_model_loader: - kv 25: tokenizer.ggml.model str = llama
llama_model_loader: - kv 26: tokenizer.ggml.pre str = default
llama_model_loader: - kv 27: tokenizer.ggml.tokens arr[str,262144] = ["<pad>", "<eos>", "<bos>", "<unk>", ...
llama_model_loader: - kv 28: tokenizer.ggml.scores arr[f32,262144] = [-1000.000000, -1000.000000, -1000.00...
llama_model_loader: - kv 29: tokenizer.ggml.token_type arr[i32,262144] = [3, 3, 3, 3, 3, 4, 3, 3, 3, 3, 3, 3, ...
llama_model_loader: - kv 30: tokenizer.ggml.bos_token_id u32 = 2
llama_model_loader: - kv 31: tokenizer.ggml.eos_token_id u32 = 1
llama_model_loader: - kv 32: tokenizer.ggml.unknown_token_id u32 = 3
llama_model_loader: - kv 33: tokenizer.ggml.padding_token_id u32 = 0
llama_model_loader: - kv 34: tokenizer.ggml.add_bos_token bool = true
llama_model_loader: - kv 35: tokenizer.ggml.add_eos_token bool = false
llama_model_loader: - kv 36: tokenizer.chat_template str = {{ bos_token }}\n{%- if messages[0]['r...
llama_model_loader: - kv 37: tokenizer.ggml.add_space_prefix bool = false
llama_model_loader: - kv 38: general.quantization_version u32 = 2
llama_model_loader: - kv 39: general.file_type u32 = 15
llama_model_loader: - type f32: 373 tensors
llama_model_loader: - type q4_K: 374 tensors
llama_model_loader: - type q6_K: 61 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q4_K - Medium
print_info: file size = 15.40 GiB (4.90 BPW)
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 6414
load: token to piece cache size = 1.9446 MB
print_info: arch = gemma3
print_info: vocab_only = 0
print_info: n_ctx_train = 131072
print_info: n_embd = 5376
print_info: n_layer = 62
print_info: n_head = 32
print_info: n_head_kv = 16
print_info: n_rot = 128
print_info: n_swa = 1024
print_info: n_swa_pattern = 6
print_info: n_embd_head_k = 128
print_info: n_embd_head_v = 128
print_info: n_gqa = 2
print_info: n_embd_k_gqa = 2048
print_info: n_embd_v_gqa = 2048
print_info: f_norm_eps = 0.0e+00
print_info: f_norm_rms_eps = 1.0e-06
print_info: f_clamp_kqv = 0.0e+00
print_info: f_max_alibi_bias = 0.0e+00
print_info: f_logit_scale = 0.0e+00
print_info: f_attn_scale = 7.7e-02
print_info: n_ff = 21504
print_info: n_expert = 0
print_info: n_expert_used = 0
print_info: causal attn = 1
print_info: pooling type = 0
print_info: rope type = 2
print_info: rope scaling = linear
print_info: freq_base_train = 1000000.0
print_info: freq_scale_train = 0.125
print_info: n_ctx_orig_yarn = 131072
print_info: rope_finetuned = unknown
print_info: ssm_d_conv = 0
print_info: ssm_d_inner = 0
print_info: ssm_d_state = 0
print_info: ssm_dt_rank = 0
print_info: ssm_dt_b_c_rms = 0
print_info: model type = 27B
print_info: model params = 27.01 B
print_info: general.name = Gemma 3 27b It
print_info: vocab type = SPM
print_info: n_vocab = 262144
print_info: n_merges = 0
print_info: BOS token = 2 '<bos>'
print_info: EOS token = 1 '<eos>'
print_info: EOT token = 106 '<end_of_turn>'
print_info: UNK token = 3 '<unk>'
print_info: PAD token = 0 '<pad>'
print_info: LF token = 248 '<0x0A>'
print_info: EOG token = 1 '<eos>'
print_info: EOG token = 106 '<end_of_turn>'
print_info: max token length = 48
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 62 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 63/63 layers to GPU
load_tensors: CUDA0 model buffer size = 15773.70 MiB
load_tensors: CPU_Mapped model buffer size = 1102.50 MiB
.........................................................................................
llama_context: constructing llama_context
llama_context: n_seq_max = 16
llama_context: n_ctx = 49152
llama_context: n_ctx_per_seq = 3072
llama_context: n_batch = 2048
llama_context: n_ubatch = 128
llama_context: causal_attn = 1
llama_context: flash_attn = 1
llama_context: freq_base = 1000000.0
llama_context: freq_scale = 0.125
llama_context: n_ctx_per_seq (3072) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_context: CUDA_Host output buffer size = 16.00 MiB
llama_kv_cache_unified_iswa: creating non-SWA KV cache, size = 49152 cells
llama_kv_cache_unified: CUDA0 KV buffer size = 3840.00 MiB
llama_kv_cache_unified: size = 3840.00 MiB ( 49152 cells, 10 layers), K (f16): 1920.00 MiB, V (f16): 1920.00 MiB
llama_kv_cache_unified_iswa: creating SWA KV cache, size = 18432 cells
llama_kv_cache_unified: CUDA0 KV buffer size = 7488.00 MiB
llama_kv_cache_unified: size = 7488.00 MiB ( 18432 cells, 52 layers), K (f16): 3744.00 MiB, V (f16): 3744.00 MiB
llama_context: CUDA0 compute buffer size = 130.62 MiB
llama_context: CUDA_Host compute buffer size = 35.63 MiB
llama_context: graph nodes = 2489
llama_context: graph splits = 2
common_init_from_params: KV cache shifting is not supported for this context, disabling KV cache shifting
llama_adapter_lora_init_impl: loading lora adapter from '/models/adapters/care_augment_river_v0.1.31.gguf' ...
llama_adapter_lora_init_impl: CUDA0 LoRA buffer size = 433.03 MiB
llama_adapter_lora_init_impl: loaded 868 tensors from lora file
common_init_from_params: setting dry_penalty_last_n to ctx_size = 49152
srv init: initializing slots, n_slots = 16
slot init: id 0 | task -1 | new slot n_ctx_slot = 3072
slot init: id 1 | task -1 | new slot n_ctx_slot = 3072
slot init: id 2 | task -1 | new slot n_ctx_slot = 3072
slot init: id 3 | task -1 | new slot n_ctx_slot = 3072
slot init: id 4 | task -1 | new slot n_ctx_slot = 3072
slot init: id 5 | task -1 | new slot n_ctx_slot = 3072
slot init: id 6 | task -1 | new slot n_ctx_slot = 3072
slot init: id 7 | task -1 | new slot n_ctx_slot = 3072
slot init: id 8 | task -1 | new slot n_ctx_slot = 3072
slot init: id 9 | task -1 | new slot n_ctx_slot = 3072
slot init: id 10 | task -1 | new slot n_ctx_slot = 3072
slot init: id 11 | task -1 | new slot n_ctx_slot = 3072
slot init: id 12 | task -1 | new slot n_ctx_slot = 3072
slot init: id 13 | task -1 | new slot n_ctx_slot = 3072
slot init: id 14 | task -1 | new slot n_ctx_slot = 3072
slot init: id 15 | task -1 | new slot n_ctx_slot = 3072
main: model loaded
main: chat template, chat_template: {{ bos_token }}
{%- if messages[0]['role'] == 'system' -%}
{%- if messages[0]['content'] is string -%}
{%- set first_user_prefix = messages[0]['content'] + '
' -%}
{%- else -%}
{%- set first_user_prefix = messages[0]['content'][0]['text'] + '
' -%}
{%- endif -%}
{%- set loop_messages = messages[1:] -%}
{%- else -%}
{%- set first_user_prefix = "" -%}
{%- set loop_messages = messages -%}
{%- endif -%}
{%- for message in loop_messages -%}
{%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
{{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
{%- endif -%}
{%- if (message['role'] == 'assistant') -%}
{%- set role = "model" -%}
{%- else -%}
{%- set role = message['role'] -%}
{%- endif -%}
{{ '<start_of_turn>' + role + '
' + (first_user_prefix if loop.first else "") }}
{%- if message['content'] is string -%}
{{ message['content'] | trim }}
{%- elif message['content'] is iterable -%}
{%- for item in message['content'] -%}
{%- if item['type'] == 'image' -%}
{{ '<start_of_image>' }}
{%- elif item['type'] == 'text' -%}
{{ item['text'] | trim }}
{%- endif -%}
{%- endfor -%}
{%- else -%}
{{ raise_exception("Invalid content type") }}
{%- endif -%}
{{ '<end_of_turn>
' }}
{%- endfor -%}
{%- if add_generation_prompt -%}
{{'<start_of_turn>model
'}}
{%- endif -%}
, example_format: '<start_of_turn>user
You are a helpful assistant
Hello<end_of_turn>
<start_of_turn>model
Hi there<end_of_turn>
<start_of_turn>user
How are you?<end_of_turn>
<start_of_turn>model
'
main: server is listening on http://0.0.0.0:8080 - starting the main loop