add kvcache fp8

504a12b8 · zhuwenwen · d074a953 · 504a12b8 · 504a12b8 · 504a12b8
Commit 504a12b8 authored Jun 12, 2025 by zhuwenwen
3 changed files
--- a/csrc/quantization/fp8/amd/quant_utils.cuh
+++ b/csrc/quantization/fp8/amd/quant_utils.cuh
--- a/csrc/quantization/fp8/amd/quant_utils_src.cuh
+++ b/csrc/quantization/fp8/amd/quant_utils_src.cuh
--- a/vllm/utils.py
+++ b/vllm/utils.py
@@ -172,7 +172,7 @@ STR_DTYPE_TO_TORCH_DTYPE = {
    "half": torch.half,
    "bfloat16": torch.bfloat16,
    "float": torch.float,
-    # "fp8": torch.uint8,
+    "fp8": torch.uint8,
    # "fp8_e4m3": torch.uint8,
    # "fp8_e5m2": torch.uint8,
    "int8": torch.int8,