Merge pull request #407 from partoneplay/main

Add support for Ollama streaming output and integrate Open-WebUI as the chat UI demo

examples/lightrag_api_open_webui_demo.py

@@ -0,0 +1,140 @@
+from datetime import datetime, timezone
+from fastapi import FastAPI
+from fastapi.responses import StreamingResponse
+import inspect
+import json
+from pydantic import BaseModel
+from typing import Optional
+
+import os
+import logging
+from lightrag import LightRAG, QueryParam
+from lightrag.llm import ollama_model_complete, ollama_embed
+from lightrag.utils import EmbeddingFunc
+
+import nest_asyncio
+
+WORKING_DIR = "./dickens"
+
+logging.basicConfig(format="%(levelname)s:%(message)s", level=logging.INFO)
+
+if not os.path.exists(WORKING_DIR):
+    os.mkdir(WORKING_DIR)
+
+rag = LightRAG(
+    working_dir=WORKING_DIR,
+    llm_model_func=ollama_model_complete,
+    llm_model_name="qwen2.5:latest",
+    llm_model_max_async=4,
+    llm_model_max_token_size=32768,
+    llm_model_kwargs={"host": "http://localhost:11434", "options": {"num_ctx": 32768}},
+    embedding_func=EmbeddingFunc(
+        embedding_dim=1024,
+        max_token_size=8192,
+        func=lambda texts: ollama_embed(
+            texts=texts, embed_model="bge-m3:latest", host="http://127.0.0.1:11434"
+        ),
+    ),
+)
+
+with open("./book.txt", "r", encoding="utf-8") as f:
+    rag.insert(f.read())
+
+# Apply nest_asyncio to solve event loop issues
+nest_asyncio.apply()
+
+app = FastAPI(title="LightRAG", description="LightRAG API open-webui")
+
+
+# Data models
+MODEL_NAME = "LightRAG:latest"
+
+
+class Message(BaseModel):
+    role: Optional[str] = None
+    content: str
+
+
+class OpenWebUIRequest(BaseModel):
+    stream: Optional[bool] = None
+    model: Optional[str] = None
+    messages: list[Message]
+
+
+# API routes
+
+
+@app.get("/")
+async def index():
+    return "Set Ollama link to http://ip:port/ollama in Open-WebUI Settings"
+
+
+@app.get("/ollama/api/version")
+async def ollama_version():
+    return {"version": "0.4.7"}
+
+
+@app.get("/ollama/api/tags")
+async def ollama_tags():
+    return {
+        "models": [
+            {
+                "name": MODEL_NAME,
+                "model": MODEL_NAME,
+                "modified_at": "2024-11-12T20:22:37.561463923+08:00",
+                "size": 4683087332,
+                "digest": "845dbda0ea48ed749caafd9e6037047aa19acfcfd82e704d7ca97d631a0b697e",
+                "details": {
+                    "parent_model": "",
+                    "format": "gguf",
+                    "family": "qwen2",
+                    "families": ["qwen2"],
+                    "parameter_size": "7.6B",
+                    "quantization_level": "Q4_K_M",
+                },
+            }
+        ]
+    }
+
+
+@app.post("/ollama/api/chat")
+async def ollama_chat(request: OpenWebUIRequest):
+    resp = rag.query(
+        request.messages[-1].content, param=QueryParam(mode="hybrid", stream=True)
+    )
+    if inspect.isasyncgen(resp):
+
+        async def ollama_resp(chunks):
+            async for chunk in chunks:
+                yield (
+                    json.dumps(
+                        {
+                            "model": MODEL_NAME,
+                            "created_at": datetime.now(timezone.utc).strftime(
+                                "%Y-%m-%dT%H:%M:%S.%fZ"
+                            ),
+                            "message": {
+                                "role": "assistant",
+                                "content": chunk,
+                            },
+                            "done": False,
+                        },
+                        ensure_ascii=False,
+                    ).encode("utf-8")
+                    + b"\n"
+                )  # the b"\n" is important
+
+        return StreamingResponse(ollama_resp(resp), media_type="application/json")
+    else:
+        return resp
+
+
+@app.get("/health")
+async def health_check():
+    return {"status": "healthy"}
+
+
+if __name__ == "__main__":
+    import uvicorn
+
+    uvicorn.run(app, host="0.0.0.0", port=8020)

examples/lightrag_ollama_demo.py

@@ -1,4 +1,6 @@
+import asyncio
 import os
+import inspect
 import logging
 from lightrag import LightRAG, QueryParam
 from lightrag.llm import ollama_model_complete, ollama_embedding
@@ -49,3 +51,20 @@
 print(
     rag.query("What are the top themes in this story?", param=QueryParam(mode="hybrid"))
 )
+
+# stream response
+resp = rag.query(
+    "What are the top themes in this story?",
+    param=QueryParam(mode="hybrid", stream=True),
+)
+
+
+async def print_stream(stream):
+    async for chunk in stream:
+        print(chunk, end="", flush=True)
+
+
+if inspect.isasyncgen(resp):
+    asyncio.run(print_stream(resp))
+else:
+    print(resp)

lightrag/base.py

@@ -19,6 +19,7 @@ class QueryParam:
     only_need_context: bool = False
     only_need_prompt: bool = False
     response_type: str = "Multiple Paragraphs"
+    stream: bool = False
     # Number of top-k items to retrieve; corresponds to entities in "local" mode and relationships in "global" mode.
     top_k: int = 60
     # Number of document chunks to retrieve.

lightrag/llm.py

@@ -4,7 +4,7 @@
 import os
 import struct
 from functools import lru_cache
-from typing import List, Dict, Callable, Any
+from typing import List, Dict, Callable, Any, Union
 
 import aioboto3
 import aiohttp
@@ -36,6 +36,13 @@
     get_best_cached_response,
 )
 
+import sys
+
+if sys.version_info < (3, 9):
+    from typing import AsyncIterator
+else:
+    from collections.abc import AsyncIterator
+
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
 
@@ -474,7 +481,8 @@ async def ollama_model_if_cache(
     system_prompt=None,
     history_messages=[],
     **kwargs,
-) -> str:
+) -> Union[str, AsyncIterator[str]]:
+    stream = True if kwargs.get("stream") else False
     kwargs.pop("max_tokens", None)
     # kwargs.pop("response_format", None) # allow json
     host = kwargs.pop("host", None)
@@ -517,28 +525,39 @@ async def ollama_model_if_cache(
                 return if_cache_return["return"]
 
     response = await ollama_client.chat(model=model, messages=messages, **kwargs)
+    if stream:
+        """ cannot cache stream response """
 
-    result = response["message"]["content"]
+        async def inner():
+            async for chunk in response:
+                yield chunk["message"]["content"]
 
-    if hashing_kv is not None:
-        await hashing_kv.upsert(
-            {
-                args_hash: {
-                    "return": result,
-                    "model": model,
-                    "embedding": quantized.tobytes().hex()
-                    if is_embedding_cache_enabled
-                    else None,
-                    "embedding_shape": quantized.shape
-                    if is_embedding_cache_enabled
-                    else None,
-                    "embedding_min": min_val if is_embedding_cache_enabled else None,
-                    "embedding_max": max_val if is_embedding_cache_enabled else None,
-                    "original_prompt": prompt,
+        return inner()
+    else:
+        result = response["message"]["content"]
+        if hashing_kv is not None:
+            await hashing_kv.upsert(
+                {
+                    args_hash: {
+                        "return": result,
+                        "model": model,
+                        "embedding": quantized.tobytes().hex()
+                        if is_embedding_cache_enabled
+                        else None,
+                        "embedding_shape": quantized.shape
+                        if is_embedding_cache_enabled
+                        else None,
+                        "embedding_min": min_val
+                        if is_embedding_cache_enabled
+                        else None,
+                        "embedding_max": max_val
+                        if is_embedding_cache_enabled
+                        else None,
+                        "original_prompt": prompt,
+                    }
                 }
-            }
-        )
-    return result
+            )
+        return result
 
 
 @lru_cache(maxsize=1)
@@ -816,7 +835,7 @@ async def hf_model_complete(
 
 async def ollama_model_complete(
     prompt, system_prompt=None, history_messages=[], keyword_extraction=False, **kwargs
-) -> str:
+) -> Union[str, AsyncIterator[str]]:
     keyword_extraction = kwargs.pop("keyword_extraction", None)
     if keyword_extraction:
         kwargs["format"] = "json"

lightrag/operate.py

@@ -534,8 +534,9 @@ async def kg_query(
     response = await use_model_func(
         query,
         system_prompt=sys_prompt,
+        stream=query_param.stream,
     )
-    if len(response) > len(sys_prompt):
+    if isinstance(response, str) and len(response) > len(sys_prompt):
         response = (
             response.replace(sys_prompt, "")
             .replace("user", "")