Quantized model repository for different backends

[shreshth-tru]

🚀 The feature, motivation and pitch

Hi,
I've been wondering if there's a public repository of .pte files for specific backends. For example, while I can generate .pte files for the Llama 3.2 1B model targeting the QNN backend, that model doesn’t work at all—it produces gibberish outputs consistently, as others have also reported.

Unfortunately, I don’t have the compute resources to generate .pte files for larger models like the 8B variant, which might actually work. This leaves the entire exercise stuck.

Is there a community repository or source where precompiled .pte files for larger models and backends are shared?

Alternatives

No response

Additional context

No response

RFC (Optional)

No response

cc @larryliu0820 @mergennachin @cccclai @helunwencser @jackzhxng

[SS-JIA]

cc: @cccclai since QNN backend was mentioned

Regarding a public .pte repository, I don't think one exists at the moment. It is a very valid suggestion, though. I will bring it up in the discord and see if there are plans to create one.

[cccclai]

Yeah it's a known issue. We will try to solve it next month

[cccclai]

We host some .pte files in https://huggingface.co/executorch-community, currently only there is only CPU versions. We will expand this effort to have more .pte files