Hass/xiaozhi-esp32-server
1
0
Fork 0
mirror of https://github.com/xinnan-tech/xiaozhi-esp32-server.git synced 2026-07-29 18:23:59 +08:00
Code Issues Packages Projects Releases Wiki Activity

本地记忆+意图识别 (#250)

Browse Source 
* 增加本地记忆功能，使用llm总结记忆

* update：增加统一非流式输出输出

* 增加意图识别内容，使用llm进行识别

* 初始化记忆模块

* 完善意图识别处理后的流程

* 通过使用function call实现意图识别

* update:优化意图识别的配置

* update:function call最优设置成doubao-pro-32k-functioncall-241028

---------

Co-authored-by: 玄凤科技 <eric230308@gmail.com>
Co-authored-by: hrz <1710360675@qq.com>
This commit is contained in:
欣南科技
2025-03-09 21:33:45 +08:00
committed by GitHub
co-authored by 玄凤科技 hrz
parent 9c2b2a2dcc
commit 63f34e5a82
19 changed files with 858 additions and 108 deletions
Download Patch File Download Diff File Expand all files Collapse all files
main/xiaozhi-server/core/providers/llm/base.py
+30
View File
@@ -1,8 +1,38 @@
from abc import ABC, abstractmethod
from config.logger import setup_logging
TAG = __name__
logger = setup_logging()
class LLMProviderBase(ABC):
@abstractmethod
def response(self, session_id, dialogue):
"""LLM response generator"""
pass
def response_no_stream(self, system_prompt, user_prompt):
try:
# 构造对话格式
dialogue = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_prompt}
]
result = ""
for part in self.response("", dialogue):
result += part
return result
except Exception as e:
logger.bind(tag=TAG).error(f"Error in Ollama response generation: {e}")
return "【LLM服务响应异常】"
def response_with_functions(self, session_id, dialogue, functions=None):
"""
Default implementation for function calling (streaming)
This should be overridden by providers that support function calls
Returns: generator that yields either text tokens or a special function call token
"""
# For providers that don't support functions, just return regular response
for token in self.response(session_id, dialogue):
yield {"type": "content", "content": token}
main/xiaozhi-server/core/providers/llm/ollama/ollama.py
+62 -27
View File
@@ -1,5 +1,6 @@
from config.logger import setup_logging
import requests, json
from openai import OpenAI
import json
from core.providers.llm.base import LLMProviderBase
TAG = __name__
@@ -8,39 +9,73 @@ logger = setup_logging()
class LLMProvider(LLMProviderBase):
def __init__(self, config):
self.model_name = config.get("model_name")
self.base_url = config.get("base_url", "http://localhost:11434")
# Initialize OpenAI client with Ollama base URL
#如果没有v1，增加v1
if not self.base_url.endswith("/v1"):
self.base_url = f"{self.base_url}/v1"
self.client = OpenAI(
base_url=self.base_url,
api_key="ollama" # Ollama doesn't need an API key but OpenAI client requires one
)
def response(self, session_id, dialogue):
def response(self, session_id, dialogue):
try:
# Convert dialogue format to Ollama format
prompt = ""
for msg in dialogue:
if msg["role"] == "system":
prompt += f"System: {msg['content']}\n"
elif msg["role"] == "user":
prompt += f"User: {msg['content']}\n"
elif msg["role"] == "assistant":
prompt += f"Assistant: {msg['content']}\n"
# Make request to Ollama API
response = requests.post(
f"{self.base_url}/api/generate",
json={
"model": self.model_name,
"prompt": prompt,
"stream": True
},
responses = self.client.chat.completions.create(
model=self.model_name,
messages=dialogue,
stream=True
)
for line in response.iter_lines():
if line:
json_response = json.loads(line)
if "response" in json_response:
yield json_response["response"]
for chunk in responses:
try:
delta = chunk.choices[0].delta if getattr(chunk, 'choices', None) else None
content = delta.content if hasattr(delta, 'content') else ''
if content:
yield content
except Exception as e:
logger.bind(tag=TAG).error(f"Error processing chunk: {e}")
except Exception as e:
logger.bind(tag=TAG).error(f"Error in Ollama response generation: {e}")
yield "【Ollama服务响应异常】"
def response_with_functions(self, session_id, dialogue, functions=None):
try:
stream = self.client.chat.completions.create(
model=self.model_name,
messages=dialogue,
stream=True,
tools=functions,
)
current_function_call = None
current_content = ""
for chunk in stream:
delta = chunk.choices[0].delta
if delta.content:
current_content += delta.content
yield {"type": "content", "content": delta.content}
if delta.tool_calls:
tool_call = delta.tool_calls[0]
# Handle the function call data using proper attribute access
if not current_function_call:
current_function_call = {
"function": {
"name": tool_call.function.name,
"arguments": tool_call.function.arguments
}
}
if current_function_call:
logger.bind(tag=TAG).debug(f"ollama Function call detected: {current_function_call}")
yield {"type": "function_call", "function_call": current_function_call}
except Exception as e:
logger.bind(tag=TAG).error(f"Error in Ollama function call: {e}")
yield {"type": "content", "content": f"【Ollama服务响应异常: {str(e)}】"}
main/xiaozhi-server/core/providers/llm/openai/openai.py
+38
View File
@@ -43,3 +43,41 @@ class LLMProvider(LLMProviderBase):
except Exception as e:
logger.bind(tag=TAG).error(f"Error in response generation: {e}")
def response_with_functions(self, session_id, dialogue, functions=None):
try:
stream = self.client.chat.completions.create(
model=self.model_name,
messages=dialogue,
stream=True,
tools=functions,
)
current_function_call = None
current_content = ""
for chunk in stream:
delta = chunk.choices[0].delta
if delta.content:
current_content += delta.content
yield {"type": "content", "content": delta.content}
if delta.tool_calls:
tool_call = delta.tool_calls[0]
# Handle the function call data using proper attribute access
if not current_function_call:
current_function_call = {
"function": {
"name": tool_call.function.name,
"arguments": tool_call.function.arguments
}
}
if current_function_call:
logger.bind(tag=TAG).debug(f"openai Function call detected: {current_function_call}")
yield {"type": "function_call", "function_call": current_function_call}
except Exception as e:
self.logger.bind(tag=TAG).error(f"Error in function call streaming: {e}")
yield {"type": "content", "content": f"【OpenAI服务响应异常: {e}】"}