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 17, 06:45 UTC
Fix OPT param names (#1819)
Woosuk Kwon committed
Nov 28, 2023 at 11:22 UTC
b9438904842b729f622f286447fc22c94bd2735f
1 file changed
+3
vllm/model_executor/models/opt.py
+3
@@ -331,6 +331,9 @@ class OPTForCausalLM(nn.Module):
331
model_name_or_path, cache_dir, load_format, revision):
332
if "lm_head.weight" in name:
333
continue
334
+ if name.startswith("decoder."):
335
+ name = "model." + name
336
+
337
for (param_name, weight_name, shard_id) in stacked_params_mapping:
338
if weight_name not in name:
339
continue