Skip to content

Fix batch embedding averaging for batch_size > 1 #3839

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Chessing234 wants to merge 2 commits into
base: main
Choose a base branch
from
+4 −3
Open

Fix batch embedding averaging for batch_size > 1 #3839

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e38008c
Fix UnboundLocalError in generate_gate when generator yields nothing
Chessing234 Apr 6, 2026
4004def
Fix batch embedding averaging for batch_size > 1
Chessing234 Apr 6, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 4 additions & 3 deletions fastchat/serve/model_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -144,6 +144,7 @@ def generate_stream_gate(self, params):
yield json.dumps(ret).encode() + b"\0"

def generate_gate(self, params):
x = b"{}\0"
for x in self.generate_stream_gate(params):
pass
return json.loads(x[:-1].decode())
Expand Down Expand Up @@ -171,7 +172,7 @@ def __process_embed_chunk(self, input_ids, attention_mask, **model_type_dict):
mask = attention_mask.unsqueeze(-1).expand(data.size()).float()
masked_embeddings = data * mask
sum_embeddings = torch.sum(masked_embeddings, dim=1)
token_num = torch.sum(attention_mask).item()
token_num = attention_mask.sum(dim=1, keepdim=True)

return sum_embeddings, token_num

Expand Down Expand Up @@ -224,7 +225,7 @@ def get_embeddings(self, params):
):
embedding = embedding / token_num
normalized_embeddings = F.normalize(embedding, p=2, dim=1)
ret["token_num"] = token_num
ret["token_num"] = token_num.sum().item()
else:
all_embeddings = []
all_token_num = 0
Expand Down Expand Up @@ -273,7 +274,7 @@ def get_embeddings(self, params):
embedding = torch.sum(all_embeddings_tensor, dim=0) / all_token_num
normalized_embeddings = F.normalize(embedding, p=2, dim=1)

ret["token_num"] = all_token_num
ret["token_num"] = all_token_num.sum().item()

if base64_encode == "base64":
out_embeddings = self.__encode_base64(normalized_embeddings)
Expand Down
Loading