A high-throughput and memory-efficient inference and serving engine for LLMs
[Bugfix][Quantization] Reject unsupported compressed tensors KV cache schemes (#45312)
Signed-off-by: Ting Sun <suntcrick@gmail.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Ting SUN committed
Jun 17, 2026 at 13:08 UTC
a46abb7ae68acc13a4fc5870db98619b3f97c6e0
1 file changed
+1
-1
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py
+1
-1
@@ -983,7 +983,7 @@ class CompressedTensorsKVCacheMethod(BaseKVCacheMethod):
983
type_ = kv_cache_scheme.get("type")
984
num_bits = kv_cache_scheme.get("num_bits")
985
986
- if type_ != "float" and num_bits != 8:
986
+ if type_ != "float" or num_bits != 8:
987
raise NotImplementedError(
988
"Currently supported kv cache quantization is "
989
"num_bits=8, type=float, however "