[fix] fix DeepGEMM blackwell input quant & ut & fix style and log (#7247)

2025-06-17 02:45:54 +08:00
parent e30ef368ab
commit 5ca07eed90
10 changed files with 285 additions and 31 deletions
--- a/python/sglang/srt/models/deepseek_v2.py
+++ b/python/sglang/srt/models/deepseek_v2.py
@@ -1829,8 +1829,10 @@ class DeepseekV2ForCausalLM(nn.Module):
                        and weight_block_size[1] == 128
                        and model_dtype == torch.bfloat16
                    ):
-                        if deep_gemm_wrapper.ENABLE_JIT_DEEPGEMM and get_bool_env_var(
-                            "SGL_USE_DEEPGEMM_BMM", "false"
+                        if (
+                            deep_gemm_wrapper.ENABLE_JIT_DEEPGEMM
+                            and not deep_gemm_wrapper.DEEPGEMM_BLACKWELL
+                            and get_bool_env_var("SGL_USE_DEEPGEMM_BMM", "false")
                        ):
                            block_scale = weight_scale
                            use_deep_gemm_bmm = True