[Bugfix][ROCm] Fix memory access fault in AITER MLA backend for DPA+FP8 KV (#47276)
Signed-off-by: simondanielsson <simon.danielsson99@hotmail.com> Co-authored-by: nnyrhila <niko.nyrhila@amd.com>
Simon Danielsson committed
Jul 6, 2026 at 23:30 UTC
482e5524fe12b2793ae3a9d861754ffd4cee10ca
1 file changed
+1
vllm/v1/attention/backends/mla/rocm_aiter_mla.py
+1
@@ -196,6 +196,7 @@ class AiterMLAMetadataBuilder(MLACommonMetadataBuilder[AiterMLAMetadata]):
196
kv_cache_dtype_str = getattr(vllm_config.cache_config, "cache_dtype", "auto")
197
if kv_cache_dtype_str in ("fp8", "fp8_e4m3", "fp8_e5m2"):
198
kv_cache_dtype_str = "fp8"
199
+ q_dtype = dtypes.fp8
200
else:
201
kv_cache_dtype_str = "bf16"
202
kv_dtype = dtypes.d_dtypes.get(kv_cache_dtype_str, dtypes.bf16)