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, 00:45 UTC
[Bugfix] Fix loading of fine-tuned models based on Phi-3-Small (#12689)
Signed-off-by: Michael Greenbaum <mgreenbaum@microsoft.com> Co-authored-by: Michael Greenbaum <mgreenbaum@microsoft.com>
Michael Greenbaum committed
Feb 4, 2025 at 14:58 UTC
6469038b149425e25b77c1ef93adf0e7712fd100
1 file changed
+2
vllm/model_executor/models/phi3_small.py
+2
@@ -476,6 +476,8 @@ class Phi3SmallForCausalLM(nn.Module, SupportsPP):
476
continue
477
if is_pp_missing_parameter(name, self):
478
continue
479
+ if "lm_head.weight" in name and self.config.tie_word_embeddings:
480
+ continue
481
param = params_dict[name]
482
weight_loader = getattr(param, "weight_loader",
483
default_weight_loader)