[ONNX] Model Test List for FX-to-ONNX Exporters for FakeTensorMode and dynamic=True

[wschin]

🚀 The feature, motivation and pitch

Some test candidates to consider

• LLaMA: https://huggingface.co/meta-llama/Llama-2-7b or https://huggingface.co/openlm-research/open_llama_7b_v2
• Whisper: https://huggingface.co/openai/whisper-large-v2
• FlanT5: https://huggingface.co/google/flan-t5-xxl
• Dolly: https://huggingface.co/databricks/dolly-v2-12b
• Falcon: https://huggingface.co/tiiuae/falcon-40b
• MPT: https://huggingface.co/mosaicml/mpt-7b
• Bloom: https://huggingface.co/bigscience/bloom

Alternatives

No response

Additional context

No response

Tasks

Beta Give feedback

1. Add ONNX export support for huggingface's bigscience/bloom-560m model #106930
   Merged ciflow/trunk open source release notes: onnx +4
2. [ONNX] Add huggingface models into CI tests #107247
   Merged ciflow/trunk module: onnx open source release notes: onnx topic: not user facing +6

[thiagocrepaldi]

I am working on bigscience/bloom-560m now

[titaiwangms]

Smaller ones

• https://huggingface.co/openlm-research/open_llama_3b_v2 (Done)
• https://huggingface.co/openai/whisper-tiny (Done)
• https://huggingface.co/google/flan-t5-small (Done)
• https://huggingface.co/databricks/dolly-v2-3b (onnx-scrip input issue)
• https://huggingface.co/tiiuae/falcon-7b (AssertionError: Mutating module attribute seq_len_cached during export.)
• https://huggingface.co/mosaicml/mpt-7b (torch._dynamo.exc.UserError: Dynamic control flow is not supported at the moment. Please use functorch.experimental.control_flow.cond to explicitly capture the control flow)
• https://huggingface.co/bigscience/bloom-560m (Done)

[titaiwangms]

1. runtime needs to run small HF models. (Use others model)
2. no runtime needs to test regular size HF models. (Use config)