A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 11, 03:45 UTC
[BugFix] Fix TypeError: unhashable type: 'dict' when serving deepseek32 (#30924)
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Lucas Wilkinson committed
Dec 19, 2025 at 16:07 UTC
5f6477d1d0b800985602a4b495e524de19782073
1 file changed
+4
-1
vllm/v1/worker/gpu_model_runner.py
+4
-1
@@ -1641,7 +1641,10 @@ class GPUModelRunner(
1641
) -> None:
1642
attn_group = self.attn_groups[kv_cache_gid][attn_gid]
1643
builder = attn_group.get_metadata_builder(ubid or 0)
1644
- cache_key = (kv_cache_groups[kv_cache_gid].kv_cache_spec, type(builder))
1644
+ kv_cache_spec = kv_cache_groups[kv_cache_gid].kv_cache_spec
1645
+ if isinstance(kv_cache_spec, UniformTypeKVCacheSpecs):
1646
+ kv_cache_spec = kv_cache_spec.kv_cache_specs[attn_group.layer_names[0]]
1647
+ cache_key = (kv_cache_spec, type(builder))
1648
1649
cascade_attn_prefix_len = (
1650
cascade_attn_prefix_lens[kv_cache_gid][attn_gid]