Skip to content

add chatglm3 conv template support in conversation.py #2622

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

Merged
infwinston merged 7 commits into from
Nov 10, 2023
Merged

add chatglm3 conv template support in conversation.py #2622

Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
8642f38
add chatglm3 conv template support in conversation.py
ZeyuTeng96 Oct 31, 2023
2183bc4
Update conversation.py
ZeyuTeng96 Nov 10, 2023
84e885f
Merge branch 'main' into patch-1
ZeyuTeng96 Nov 10, 2023
c92675f
Update model_adapter.py to support chatglm3
ZeyuTeng96 Nov 10, 2023
dcd5324
Update model_adapter.py
ZeyuTeng96 Nov 10, 2023
efd1501
Update model_adapter.py
ZeyuTeng96 Nov 10, 2023
99087e7
Update conversation.py
ZeyuTeng96 Nov 10, 2023
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
26 changes: 26 additions & 0 deletions fastchat/conversation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,7 @@ class SeparatorStyle(IntEnum):
PHOENIX = auto()
ROBIN = auto()
FALCON_CHAT = auto()
CHATGLM3 = auto()


@dataclasses.dataclass
Expand Down Expand Up @@ -163,6 +164,14 @@ def get_prompt(self) -> str:
else:
ret += role + "\n"
return ret
elif self.sep_style == SeparatorStyle.CHATGLM3:
Copy link
Member

@infwinston infwinston Nov 2, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you add reference?

Copy link
Contributor Author

@ZeyuTeng96 ZeyuTeng96 Nov 3, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How about adding this ref (https://huggingface.co/THUDM/chatglm3-6b/blob/fc3235f807ef5527af598c05f04f2ffd17f48bab/tokenization_chatglm.py#L184) ?

Copy link

@rainbend rainbend Nov 3, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ZeyuTeng96 Like this conversation.py#L167 :)

ret = "" if system_prompt == "" else system_prompt
Copy link
Contributor

@Jeffwan Jeffwan Nov 5, 2023 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

need \n or not?

Copy link

@duzx16 duzx16 Nov 9, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It doesn't need

for role, message in self.messages:
if message:
ret += role + "\n" + message
Copy link
Contributor

@Jeffwan Jeffwan Nov 5, 2023 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same here. need \n ending or not?

Copy link

@duzx16 duzx16 Nov 9, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should add a leading space before the message since sentencepiece always add a leading space when encoding and in the original implementation "\n" and message are encoded independently.

Line 171 should be

ret += role + "\n" + " " + message

else:
ret += role
return ret
elif self.sep_style == SeparatorStyle.CHATINTERN:
# source: https://huggingface.co/internlm/internlm-chat-7b-8k/blob/bd546fa984b4b0b86958f56bf37f94aa75ab8831/modeling_internlm.py#L771
seps = [self.sep, self.sep2]
Expand Down Expand Up @@ -448,6 +457,23 @@ def get_conv_template(name: str) -> Conversation:
)
)

# ChatGLM3 default template
register_conv_template(
Conversation(
name="chatglm3",
system_template="<|system|>\n{system_message}",
system_message="You are ChatGLM3, a large language model trained by Zhipu.AI. Follow the user's instructions carefully. Respond using markdown.",
roles=("<|user|>", "<|assistant|>"),
sep_style=SeparatorStyle.CHATGLM3,
sep="",
stop_token_ids=[
64795,
64797,
2,
], # "<|user|>", "<|observation|>", "</s>"
)
)

# Dolly V2 default template
register_conv_template(
Conversation(
Expand Down