github/OpenHands
1
0
Fork 0
mirror of https://github.com/OpenHands/OpenHands.git synced 2026-03-22 13:47:19 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat(agent): CodeAct with function calling (#4537)

Browse Source 
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: tobitege <10787084+tobitege@users.noreply.github.com>
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
Co-authored-by: tofarr <tofarr@gmail.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
This commit is contained in:
Xingyao Wang
2024-10-28 22:06:33 -05:00
committed by GitHub
parent 421b4c108a
commit ae13171194
34 changed files with 1834 additions and 235 deletions
Download Patch File Download Diff File Expand all files Collapse all files

59
openhands/core/message.py
View File

@@ -1,6 +1,7 @@
from enum import Enum
from typing import Literal
from litellm import ChatCompletionMessageToolCall
from pydantic import BaseModel, Field, model_serializer
@@ -48,10 +49,16 @@ class ImageContent(Content):
class Message(BaseModel):
role: Literal['user', 'system', 'assistant']
content: list[TextContent | ImageContent] = Field(default=list)
role: Literal['user', 'system', 'assistant', 'tool']
content: list[TextContent | ImageContent] = Field(default_factory=list)
cache_enabled: bool = False
vision_enabled: bool = False
# function calling
# - tool calls (from LLM)
tool_calls: list[ChatCompletionMessageToolCall] | None = None
# - tool execution result (to LLM)
tool_call_id: str | None = None
name: str | None = None # name of the tool
@property
def contains_image(self) -> bool:
@@ -59,23 +66,31 @@ class Message(BaseModel):
@model_serializer
def serialize_model(self) -> dict:
content: list[dict] | str
# two kinds of serializer:
# 1. vision serializer: when prompt caching or vision is enabled
# 2. single text serializer: for other cases
# remove this when liteLLM or providers support this format translation
if self.cache_enabled or self.vision_enabled:
# when prompt caching or vision is enabled, use vision serializer
content = []
for item in self.content:
if isinstance(item, TextContent):
content.append(item.model_dump())
elif isinstance(item, ImageContent):
content.extend(item.model_dump())
else:
# for other cases, concatenate all text content
# into a single string per message
content = '\n'.join(
item.text for item in self.content if isinstance(item, TextContent)
)
return {'content': content, 'role': self.role}
content: list[dict] = []
role_tool_with_prompt_caching = False
for item in self.content:
d = item.model_dump()
# We have to remove cache_prompt for tool content and move it up to the message level
# See discussion here for details: https://github.com/BerriAI/litellm/issues/6422#issuecomment-2438765472
if self.role == 'tool' and item.cache_prompt:
role_tool_with_prompt_caching = True
d.pop('cache_control')
if isinstance(item, TextContent):
content.append(d)
elif isinstance(item, ImageContent) and self.vision_enabled:
content.extend(d)
ret: dict = {'content': content, 'role': self.role}
if role_tool_with_prompt_caching:
ret['cache_control'] = {'type': 'ephemeral'}
if self.tool_call_id is not None:
assert (
self.name is not None
), 'name is required when tool_call_id is not None'
ret['tool_call_id'] = self.tool_call_id
ret['name'] = self.name
if self.tool_calls:
ret['tool_calls'] = self.tool_calls
return ret