forked from PaddlePaddle/FastDeploy
-
Notifications
You must be signed in to change notification settings - Fork 0
Open
Description
Add unit tests for masked_per_token_quant.
⚠️ Important: Investigate before implementing
Three upstream tasks (PaddlePaddle#20, PaddlePaddle#35, PaddlePaddle#51) reference masked_per_token_quant. Tasks PaddlePaddle#20 and PaddlePaddle#35 are COMPLETED (PRs #4111 and #3867).
This task (PaddlePaddle#51) must cover something NOT already tested.
The op masked_per_token_quant does NOT appear directly in cpp_extensions.cc. Check:
per_token_quantatcpp_extensions.cc:1283test_fused_masked_swiglu_quant.pyfor the fused version- Python modules for possible pure-Python implementation
Source files to study:
custom_ops/gpu_ops/cpp_extensions.cc— searchper_token_quanttests/operators/test_fused_masked_swiglu_quant.py— reference for fused versiontests/operators/test_dynamic_per_token_scaled_fp8_quant.py— related quant test
Test file: tests/operators/test_masked_per_token_quant.py
Branch: task/051-masked-per-token-quant-test
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels