Skip to content

GitLab

Commit a9c755ac authored by zhuwenwen's avatar zhuwenwen
Browse files

update kv_cache_dtype

parent c0697921
Show whitespace changes
Inline Side-by-side
Showing with 1 addition and 1 deletion
vllm/v1/attention/backends/mla/common.py
View file @ a9c755ac
...@@ -1300,7 +1300,7 @@ class MLACommonImpl(MLAAttentionImpl[M], Generic[M]): ...@@ -1300,7 +1300,7 @@ class MLACommonImpl(MLAAttentionImpl[M], Generic[M]):
if has_decode: if has_decode:
assert attn_metadata.decode is not None assert attn_metadata.decode is not None
if torch.cuda.get_device_properties("cuda").gcnArchName.split(':')[0] == "gfx938" and kv_cache_dtype_str=="fp8_e4m3" and envs.VLLM_USE_FUSED_CACHE_QUANT_BMM_MLA: if torch.cuda.get_device_properties("cuda").gcnArchName.split(':')[0] == "gfx938" and self.kv_cache_dtype=="fp8_e4m3" and envs.VLLM_USE_FUSED_CACHE_QUANT_BMM_MLA:
decode_q = q_quant[:num_decode_tokens] decode_q = q_quant[:num_decode_tokens]
decode_q_nope, decode_q_pe = decode_q.split( decode_q_nope, decode_q_pe = decode_q.split(
[self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1) [self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment