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 16, 08:15 UTC
[4.5/N] bugfix for quant config in speculative decode (#10007)
Signed-off-by: youkaichao <youkaichao@gmail.com>
youkaichao committed
Nov 4, 2024 at 15:11 UTC
2094062b4eafe465826e936fbd5cbd8f099d7762
1 file changed
+4
vllm/spec_decode/spec_decode_worker.py
+4
@@ -61,6 +61,10 @@ def create_spec_worker(*args, **kwargs) -> "SpecDecodeWorker":
61
62
draft_worker_config = copy.deepcopy(vllm_config)
63
draft_worker_config.model_config = speculative_config.draft_model_config
64
+ draft_worker_config.quant_config = VllmConfig._get_quantization_config(
65
+ draft_worker_config.model_config,
66
+ vllm_config.load_config,
67
+ )
68
draft_worker_config.parallel_config = speculative_config.draft_parallel_config # noqa
69
# TODO allow draft-model specific load config.
70