xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-07 07:12:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Misc] Fix quantization-related typos (#31116)

Browse Source 
Signed-off-by: c0de128 <kevin.mckay@outlook.com>
This commit is contained in:
Kevin McKay 2025-12-21 23:13:48 -06:00 committed by GitHub
parent 8c084de59d
commit ec58c10ce1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
tests/kernels/moe/modular_kernel_tools/common.py
View File

@ -258,16 +258,16 @@ class Config:
f"{self.fe_supported_types()}." f"{self.fe_supported_types()}."
) )
# Check block quanization support # Check block quantization support
is_block_quatized = self.quant_block_shape is not None is_block_quantized = self.quant_block_shape is not None
if is_block_quatized and self.quant_dtype is None: if is_block_quantized and self.quant_dtype is None:
return False, "No block quantization support." return False, "No block quantization support."
if is_block_quatized and not self.is_block_quant_supported(): if is_block_quantized and not self.is_block_quant_supported():
return False, "Mismatched block quantization support." return False, "Mismatched block quantization support."
# deep_gemm only works with block-quantized # deep_gemm only works with block-quantized
if self.needs_deep_gemm() and not is_block_quatized: if self.needs_deep_gemm() and not is_block_quantized:
return False, "Needs DeepGEMM but not block quantized." return False, "Needs DeepGEMM but not block quantized."
# Check dependencies (turn into asserts?) # Check dependencies (turn into asserts?)

2
tests/quantization/test_fp8.py
View File

@ -217,7 +217,7 @@ def test_scaled_fp8_quant(dtype) -> None:
ref_y, inv_scale = ops.scaled_fp8_quant(x, None) ref_y, inv_scale = ops.scaled_fp8_quant(x, None)
ref_y = per_tensor_dequantize(ref_y, inv_scale, dtype) ref_y = per_tensor_dequantize(ref_y, inv_scale, dtype)
# Reference dynamic quantizaton # Reference dynamic quantization
y = quantize_ref(x, inv_scale) y = quantize_ref(x, inv_scale)
torch.testing.assert_close(ref_y, per_tensor_dequantize(y, inv_scale, dtype)) torch.testing.assert_close(ref_y, per_tensor_dequantize(y, inv_scale, dtype))

2
vllm/utils/deep_gemm.py
View File

@ -389,7 +389,7 @@ def should_use_deepgemm_for_fp8_linear(
# Verify DeepGEMM N/K dims requirements # Verify DeepGEMM N/K dims requirements
# NOTE: Also synchronized with test_w8a8_block_fp8_deep_gemm_matmul # NOTE: Also synchronized with test_w8a8_block_fp8_deep_gemm_matmul
# test inside kernels/quatization/test_block_fp8.py # test inside kernels/quantization/test_block_fp8.py
N_MULTIPLE = 64 N_MULTIPLE = 64
K_MULTIPLE = 128 K_MULTIPLE = 128