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 8, 08:45 UTC
[ROCm][quantization] improve OCP weight quant parser robust (#34431)
Signed-off-by: xuebwang-amd <xuebwang@amd.com> Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
xuebwang-amd committed
Feb 13, 2026 at 01:40 UTC
766e1678210d797757dcfe28f05184a251685dfe
1 file changed
+7
vllm/model_executor/layers/quantization/quark/quark.py
+7
@@ -337,6 +337,13 @@ class QuarkConfig(QuantizationConfig):
337
)
338
return False
339
340
+ if isinstance(weight_quant, list):
341
+ logger.debug(
342
+ "Quark model's weight quantization is incompatible with OCP_MX format: "
343
+ "weight_quant is a list (e.g. fp8_w4a8), OCP_MX requires a single dict."
344
+ )
345
+ return False
346
+
347
# Input and weight qscheme needs to be per group.
348
if weight_quant.get("qscheme") != "per_group":
349
logger.debug(