xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 07:57:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

expand coverage of gpt2 model loading (#271)

Browse Source
This commit is contained in:
twaka 2023-06-27 22:27:41 +09:00 committed by GitHub
parent 43710e8d09
commit 4026a049d3
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/gpt2.py
View File

@ -228,10 +228,12 @@ class GPT2LMHeadModel(nn.Module):
# GPT-2 ties the weights of the embedding layer and the final # GPT-2 ties the weights of the embedding layer and the final
# linear layer. # linear layer.
continue continue
if ".attn.bias" in name: if ".attn.bias" in name or ".attn.masked_bias" in name:
# Skip attention mask. # Skip attention mask.
# NOTE: "c_attn.bias" should not be skipped. # NOTE: "c_attn.bias" should not be skipped.
continue continue
if not name.startswith("transformer."):
name = "transformer." + name name = "transformer." + name
# The HF's GPT-2 implementation uses Conv1D instead of Linear. # The HF's GPT-2 implementation uses Conv1D instead of Linear.