Add torchao checkpoint tests #14074
Add torchao checkpoint tests #14074
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/14074
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 1 Cancelled Job, 37 PendingAs of commit 2f47f54 with merge base a90e907 ( NEW FAILURES - The following jobs have failed:
CANCELLED JOB - The following job was cancelled. Please retry:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
metascroy
added
ciflow/trunk
and removed
CLA Signed
This PR needs a
|
meta-cla
bot
added
the
CLA Signed
| case "$MODEL_NAME" in | ||
| qwen3_4b) | ||
| echo "Running Qwen3-4B export..." | ||
| HF_MODEL_DIR=$(hf download metascroy/Qwen3-4B-INT8-INT4) |
There was a problem hiding this comment.
TODO: before landing, update the PyTorch checkpoint and change this to pytorch/Qwen3-4B-INT8-INT4
|
|
||
| phi_4_mini) | ||
| echo "Running Phi-4-mini export..." | ||
| HF_MODEL_DIR=$(hf download metascroy/Phi-4-mini-instruct-INT8-INT4) |
There was a problem hiding this comment.
TODO: before landing, update the PyTorch checkpoint and change this to pytorch/Phi-4-mini-instruct-INT8-INT4
| cmake -DPYTHON_EXECUTABLE=python \ | ||
| -DCMAKE_INSTALL_PREFIX=cmake-out \ | ||
| -DEXECUTORCH_ENABLE_LOGGING=1 \ | ||
| -DCMAKE_BUILD_TYPE=Release \ | ||
| -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \ | ||
| -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \ | ||
| -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \ | ||
| -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \ | ||
| -DEXECUTORCH_BUILD_XNNPACK=ON \ | ||
| -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \ | ||
| -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \ | ||
| -DEXECUTORCH_BUILD_EXTENSION_LLM_RUNNER=ON \ | ||
| -DEXECUTORCH_BUILD_EXTENSION_LLM=ON \ | ||
| -DEXECUTORCH_BUILD_KERNELS_LLM=ON \ | ||
| -Bcmake-out . | ||
| cmake --build cmake-out -j16 --config Release --target install |
There was a problem hiding this comment.
shall we just test via the preset now?
cmake --preset llm -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=cmake-out
There was a problem hiding this comment.
This doesn't seem to work. I'm reverting back to not using the preset.
I filed an issue here: #14132
| @@ -0,0 +1,139 @@ | |||
| #!/usr/bin/env bash | |||
| set -euo pipefail | |||
There was a problem hiding this comment.
please use -x as well for scripts that will run in CI; makes debugging easier
This PR adds new tests that the pre-quantized model checkpoints we publish on pytorch work with ExecuTorch (lowering and C++ runner).
qwen3-4b is tested for both lowering and runtime.
phi4-mini is tested for lowering. There appears to be a regression in the C++ HF tokenizer used in the ExecuTorch, and it no longer works with the phi4-mini tokenizer. See #14077