huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-11 20:34:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix Phi-2 with tp>1

Browse Source 
We were using the wrong parallelism in the up-projection.
This commit is contained in:
Daniël de Kok 2024-06-04 08:25:33 +00:00
parent df71aafdcc
commit f9c354d120
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
server/text_generation_server/models/custom_modeling/flash_phi_modeling.py
View File

@ -238,7 +238,7 @@ class PhiMLP(nn.Module):
)
# llama weights are up_proj and down_proj and bias=False
self.up_proj = TensorParallelRowLinear.load(
self.up_proj = TensorParallelColumnLinear.load(
config,
prefix=f"{prefix}.fc1",
weights=weights,