update fp8 native implementation

7a81bc31 · zhuwenwen · 98a011e9 · 7a81bc31 · 7a81bc31
Commit 7a81bc31 authored Dec 02, 2025 by zhuwenwen
Expand all Hide whitespace changes
Inline Side-by-side

Showing with 402 additions and 366 deletions

csrc/quantization/fp8/amd/quant_utils.cuh csrc/quantization/fp8/amd/quant_utils.cuh +400 -364

vllm/utils/__init__.py vllm/utils/__init__.py +2 -2

No files found.
--- a/csrc/quantization/fp8/amd/quant_utils.cuh
+++ b/csrc/quantization/fp8/amd/quant_utils.cuh
--- a/vllm/utils/__init__.py
+++ b/vllm/utils/__init__.py
@@ -132,8 +132,8 @@ STR_DTYPE_TO_TORCH_DTYPE = {
    "bfloat16": torch.bfloat16,
    "float": torch.float,
    "fp8": torch.uint8,
-    # "fp8_e4m3": torch.uint8,
-    # "fp8_e5m2": torch.uint8,
+    "fp8_e4m3": torch.uint8,
+    "fp8_e5m2": torch.uint8,
    "int8": torch.int8,
    "fp8_inc": torch.float8_e4m3fn,
    "fp8_ds_mla": torch.uint8,