estel/dify
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

fix(api/core/model_runtime/model_providers/baichuan,localai): Parse ToolPromptMessage. #4943 (#5138)

Browse Source 
Co-authored-by: -LAN- <laipz8200@outlook.com>
This commit is contained in:
yanghx
2024-06-13 05:08:30 +00:00
committed by GitHub
parent 742b08e1d5
commit adc948e87c
2 changed files with 88 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
api/core/model_runtime/model_providers/baichuan/llm/llm.py
View File

@@ -7,6 +7,7 @@ from core.model_runtime.entities.message_entities import (
PromptMessage,
PromptMessageTool,
SystemPromptMessage,
ToolPromptMessage,
UserPromptMessage,
)
from core.model_runtime.errors.invoke import (
@@ -38,14 +39,15 @@ class BaichuanLarguageModel(LargeLanguageModel):
stream: bool = True, user: str | None = None) \
-> LLMResult | Generator:
return self._generate(model=model, credentials=credentials, prompt_messages=prompt_messages,
model_parameters=model_parameters, tools=tools, stop=stop, stream=stream, user=user)
model_parameters=model_parameters, tools=tools, stop=stop, stream=stream, user=user)
def get_num_tokens(self, model: str, credentials: dict, prompt_messages: list[PromptMessage],
tools: list[PromptMessageTool] | None = None) -> int:
return self._num_tokens_from_messages(prompt_messages)
def _num_tokens_from_messages(self, messages: list[PromptMessage],) -> int:
def _num_tokens_from_messages(self, messages: list[PromptMessage], ) -> int:
"""Calculate num tokens for baichuan model"""
def tokens(text: str):
return BaichuanTokenizer._get_num_tokens(text)
@@ -85,6 +87,17 @@ class BaichuanLarguageModel(LargeLanguageModel):
elif isinstance(message, SystemPromptMessage):
message = cast(SystemPromptMessage, message)
message_dict = {"role": "user", "content": message.content}
elif isinstance(message, ToolPromptMessage):
# copy from core/model_runtime/model_providers/anthropic/llm/llm.py
message = cast(ToolPromptMessage, message)
message_dict = {
"role": "user",
"content": [{
"type": "tool_result",
"tool_use_id": message.tool_call_id,
"content": message.content
}]
}
else:
raise ValueError(f"Unknown message type {type(message)}")
@@ -107,8 +120,8 @@ class BaichuanLarguageModel(LargeLanguageModel):
raise CredentialsValidateFailedError(f"Invalid API key: {e}")
def _generate(self, model: str, credentials: dict, prompt_messages: list[PromptMessage],
model_parameters: dict, tools: list[PromptMessageTool] | None = None,
stop: list[str] | None = None, stream: bool = True, user: str | None = None) \
model_parameters: dict, tools: list[PromptMessageTool] | None = None,
stop: list[str] | None = None, stream: bool = True, user: str | None = None) \
-> LLMResult | Generator:
if tools is not None and len(tools) > 0:
raise InvokeBadRequestError("Baichuan model doesn't support tools")
@@ -129,7 +142,8 @@ class BaichuanLarguageModel(LargeLanguageModel):
]
# invoke model
response = instance.generate(model=model, stream=stream, messages=messages, parameters=model_parameters, timeout=60)
response = instance.generate(model=model, stream=stream, messages=messages, parameters=model_parameters,
timeout=60)
if stream:
return self._handle_chat_generate_stream_response(model, prompt_messages, credentials, response)
@@ -141,7 +155,9 @@ class BaichuanLarguageModel(LargeLanguageModel):
credentials: dict,
response: BaichuanMessage) -> LLMResult:
# convert baichuan message to llm result
usage = self._calc_response_usage(model=model, credentials=credentials, prompt_tokens=response.usage['prompt_tokens'], completion_tokens=response.usage['completion_tokens'])
usage = self._calc_response_usage(model=model, credentials=credentials,
prompt_tokens=response.usage['prompt_tokens'],
completion_tokens=response.usage['completion_tokens'])
return LLMResult(
model=model,
prompt_messages=prompt_messages,
@@ -158,7 +174,9 @@ class BaichuanLarguageModel(LargeLanguageModel):
response: Generator[BaichuanMessage, None, None]) -> Generator:
for message in response:
if message.usage:
usage = self._calc_response_usage(model=model, credentials=credentials, prompt_tokens=message.usage['prompt_tokens'], completion_tokens=message.usage['completion_tokens'])
usage = self._calc_response_usage(model=model, credentials=credentials,
prompt_tokens=message.usage['prompt_tokens'],
completion_tokens=message.usage['completion_tokens'])
yield LLMResultChunk(
model=model,
prompt_messages=prompt_messages,

42
api/core/model_runtime/model_providers/localai/llm/llm.py
View File

@@ -27,6 +27,7 @@ from core.model_runtime.entities.message_entities import (
PromptMessage,
PromptMessageTool,
SystemPromptMessage,
ToolPromptMessage,
UserPromptMessage,
)
from core.model_runtime.entities.model_entities import (
@@ -57,7 +58,7 @@ class LocalAILanguageModel(LargeLanguageModel):
stream: bool = True, user: str | None = None) \
-> LLMResult | Generator:
return self._generate(model=model, credentials=credentials, prompt_messages=prompt_messages,
model_parameters=model_parameters, tools=tools, stop=stop, stream=stream, user=user)
model_parameters=model_parameters, tools=tools, stop=stop, stream=stream, user=user)
def get_num_tokens(self, model: str, credentials: dict, prompt_messages: list[PromptMessage],
tools: list[PromptMessageTool] | None = None) -> int:
@@ -69,6 +70,7 @@ class LocalAILanguageModel(LargeLanguageModel):
Calculate num tokens for baichuan model
LocalAI does not supports
"""
def tokens(text: str):
"""
We cloud not determine which tokenizer to use, cause the model is customized.
@@ -133,6 +135,7 @@ class LocalAILanguageModel(LargeLanguageModel):
:param tools: tools for tool calling
:return: number of tokens
"""
def tokens(text: str):
return self._get_num_tokens_by_gpt2(text)
@@ -247,8 +250,8 @@ class LocalAILanguageModel(LargeLanguageModel):
return entity
def _generate(self, model: str, credentials: dict, prompt_messages: list[PromptMessage],
model_parameters: dict, tools: list[PromptMessageTool] | None = None,
stop: list[str] | None = None, stream: bool = True, user: str | None = None) \
model_parameters: dict, tools: list[PromptMessageTool] | None = None,
stop: list[str] | None = None, stream: bool = True, user: str | None = None) \
-> LLMResult | Generator:
kwargs = self._to_client_kwargs(credentials)
@@ -351,6 +354,17 @@ class LocalAILanguageModel(LargeLanguageModel):
elif isinstance(message, SystemPromptMessage):
message = cast(SystemPromptMessage, message)
message_dict = {"role": "system", "content": message.content}
elif isinstance(message, ToolPromptMessage):
# copy from core/model_runtime/model_providers/anthropic/llm/llm.py
message = cast(ToolPromptMessage, message)
message_dict = {
"role": "user",
"content": [{
"type": "tool_result",
"tool_use_id": message.tool_call_id,
"content": message.content
}]
}
else:
raise ValueError(f"Unknown message type {type(message)}")
@@ -377,10 +391,10 @@ class LocalAILanguageModel(LargeLanguageModel):
return prompts
def _handle_completion_generate_response(self, model: str,
prompt_messages: list[PromptMessage],
credentials: dict,
response: Completion,
) -> LLMResult:
prompt_messages: list[PromptMessage],
credentials: dict,
response: Completion,
) -> LLMResult:
"""
Handle llm chat response
@@ -407,7 +421,8 @@ class LocalAILanguageModel(LargeLanguageModel):
)
completion_tokens = self._num_tokens_from_messages(messages=[assistant_prompt_message], tools=[])
usage = self._calc_response_usage(model=model, credentials=credentials, prompt_tokens=prompt_tokens, completion_tokens=completion_tokens)
usage = self._calc_response_usage(model=model, credentials=credentials, prompt_tokens=prompt_tokens,
completion_tokens=completion_tokens)
response = LLMResult(
model=model,
@@ -452,7 +467,8 @@ class LocalAILanguageModel(LargeLanguageModel):
prompt_tokens = self._num_tokens_from_messages(messages=prompt_messages, tools=tools)
completion_tokens = self._num_tokens_from_messages(messages=[assistant_prompt_message], tools=tools)
usage = self._calc_response_usage(model=model, credentials=credentials, prompt_tokens=prompt_tokens, completion_tokens=completion_tokens)
usage = self._calc_response_usage(model=model, credentials=credentials, prompt_tokens=prompt_tokens,
completion_tokens=completion_tokens)
response = LLMResult(
model=model,
@@ -465,10 +481,10 @@ class LocalAILanguageModel(LargeLanguageModel):
return response
def _handle_completion_generate_stream_response(self, model: str,
prompt_messages: list[PromptMessage],
credentials: dict,
response: Stream[Completion],
tools: list[PromptMessageTool]) -> Generator:
prompt_messages: list[PromptMessage],
credentials: dict,
response: Stream[Completion],
tools: list[PromptMessageTool]) -> Generator:
full_response = ''
for chunk in response: