feat: PyTorch PT2 Model Generation #8547
Conversation
qa/common/gen_qa_models.py
Outdated
| # Use a different model name for the non-batching variant | ||
| model_name = tu.get_model_name( | ||
| "libtorch_nobatch" if max_batch == 0 else "libtorch", | ||
| "libtorch_nobatch" if max_batch == 0 else "libtorch2" if torch2 else "libtorch", |
There was a problem hiding this comment.
Please refactor and make the statement clear
There was a problem hiding this comment.
done. also, apologies for the force push, hadn't realized that anyone had reviewed the changes already. won't happen again. thanks.
qa/common/gen_qa_torch_aoti_model.py
Outdated
| with open(f"{config_path}/config.pbtxt", "w") as f: | ||
| f.write( | ||
| """ | ||
| backend: "pytorch" |
There was a problem hiding this comment.
trying to keep them as the same backend.
There was a problem hiding this comment.
I see. They differs by "platform"?
There was a problem hiding this comment.
Correct. That's the plan. Ideally we don't have competing "pytorch" and "pytorch_inductor" backends.
There was a problem hiding this comment.
renamed the platform to torch_aoti to better reflect what it is and does.
whoisj
force-pushed
the
jwyman/pt2/gen-models
branch
from
55ebafa to
558c361
Compare
This change adds the generation of PT2 format serialized PyTorch model files to the QA model generation scripts.
whoisj
force-pushed
the
jwyman/pt2/gen-models
branch
from
558c361 to
78f755f
Compare
| TEST_RESULT_FILE="./test_results.txt" | ||
| CLIENT_PY="./python_based_backends_test.py" | ||
| GEN_PYTORCH_MODEL_PY="../../common/gen_qa_pytorch_model.py" | ||
| GEN_TORCHAOTI_MODEL_PY="../../common/gen_qa_torch_aoti_model.py" |
There was a problem hiding this comment.
Where is this variable used?
qa/common/gen_qa_torch_aoti_model.py
Outdated
| @@ -0,0 +1,124 @@ | |||
| # Copyright 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved. | |||
There was a problem hiding this comment.
Where is this file being used?
whoisj
force-pushed
the
jwyman/pt2/gen-models
branch
2 times, most recently
from
5dd1300 to
53b28e7
Compare
whoisj
force-pushed
the
jwyman/pt2/gen-models
branch
from
53b28e7 to
16c6c0a
Compare
This change adds the generation of PT2 format serialized PyTorch model files to the QA model generation scripts.
TRI-172