A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 11, 03:45 UTC
[Misc] Fix quantization-related typos (#31116)
Signed-off-by: c0de128 <kevin.mckay@outlook.com>
Kevin McKay committed
Dec 21, 2025 at 23:13 UTC
ec58c10ce1554eddca6c5c0eaa93ebeb4ed024d1
3 files changed
+7
-7
tests/kernels/moe/modular_kernel_tools/common.py
+5
-5
@@ -258,16 +258,16 @@ class Config:
258
f"{self.fe_supported_types()}."
259
)
260
261
- # Check block quanization support
262
- is_block_quatized = self.quant_block_shape is not None
263
- if is_block_quatized and self.quant_dtype is None:
261
+ # Check block quantization support
262
+ is_block_quantized = self.quant_block_shape is not None
263
+ if is_block_quantized and self.quant_dtype is None:
264
return False, "No block quantization support."
265
266
- if is_block_quatized and not self.is_block_quant_supported():
266
+ if is_block_quantized and not self.is_block_quant_supported():
267
return False, "Mismatched block quantization support."
268
269
# deep_gemm only works with block-quantized
270
- if self.needs_deep_gemm() and not is_block_quatized:
270
+ if self.needs_deep_gemm() and not is_block_quantized:
271
return False, "Needs DeepGEMM but not block quantized."
272
273
# Check dependencies (turn into asserts?)
tests/quantization/test_fp8.py
+1
-1
@@ -217,7 +217,7 @@ def test_scaled_fp8_quant(dtype) -> None:
217
ref_y, inv_scale = ops.scaled_fp8_quant(x, None)
218
ref_y = per_tensor_dequantize(ref_y, inv_scale, dtype)
219
220
- # Reference dynamic quantizaton
220
+ # Reference dynamic quantization
221
y = quantize_ref(x, inv_scale)
222
torch.testing.assert_close(ref_y, per_tensor_dequantize(y, inv_scale, dtype))
223
vllm/utils/deep_gemm.py
+1
-1
@@ -389,7 +389,7 @@ def should_use_deepgemm_for_fp8_linear(
389
390
# Verify DeepGEMM N/K dims requirements
391
# NOTE: Also synchronized with test_w8a8_block_fp8_deep_gemm_matmul
392
- # test inside kernels/quatization/test_block_fp8.py
392
+ # test inside kernels/quantization/test_block_fp8.py
393
N_MULTIPLE = 64
394
K_MULTIPLE = 128
395