Skip to content

Commit

Permalink
fix marlin flag set on hpu (#540)
Browse files Browse the repository at this point in the history
  • Loading branch information
nirda7 authored Nov 25, 2024
1 parent 39c6b6c commit 5eb8b1f
Showing 1 changed file with 6 additions and 4 deletions.
10 changes: 6 additions & 4 deletions vllm/model_executor/layers/quantization/fp8.py
Original file line number Diff line number Diff line change
Expand Up @@ -120,10 +120,12 @@ def __init__(self, quant_config: Fp8Config):
if current_platform.is_cuda_alike():
self.cutlass_fp8_supported = cutlass_fp8_supported()

# For GPUs that lack FP8 hardware support, we can leverage the Marlin
# kernel for fast weight-only FP8 quantization
self.use_marlin = (not current_platform.has_device_capability(89)
or envs.VLLM_TEST_FORCE_FP8_MARLIN)
self.use_marlin = False
if not current_platform.is_hpu():
# For GPUs that lack FP8 hardware support, we can leverage the
# Marlin kernel for fast weight-only FP8 quantization
self.use_marlin = (not current_platform.has_device_capability(89)
or envs.VLLM_TEST_FORCE_FP8_MARLIN)
# Disable marlin for rocm
if current_platform.is_rocm():
self.use_marlin = False
Expand Down

0 comments on commit 5eb8b1f

Please sign in to comment.