Spaces:
Sleeping
Sleeping
Refactor: Implement streaming response and simplify architecture
Browse files- Replace LangGraph with a direct Gradio implementation for simplicity and performance.
- Implement streaming responses using `TextIteratorStreamer` for a better user experience.
- Use `tokenizer.apply_chat_template` for robust prompt formatting.
- Remove obsolete `graph.py.
GEMINI.md
CHANGED
|
@@ -19,21 +19,20 @@
|
|
| 19 |
|
| 20 |
# 子目标
|
| 21 |
## 未完成
|
| 22 |
-
- [ ] **(进行中)**
|
| 23 |
-
- [ ] (已暂停) 实现自动化部署和验证流程。
|
| 24 |
|
| 25 |
## 已完成
|
|
|
|
| 26 |
- [x] 使用 LangGraph 实现一个可以路由两个模型的聊天网页应用。
|
| 27 |
|
| 28 |
---
|
| 29 |
|
| 30 |
# Todolist
|
| 31 |
## 未完成
|
| 32 |
-
- [ ] **当前任务**: 修改 `app.py`,移除 `Ling-flash-2.0` 模型,只保留 `Ring-mini-2.0`。
|
| 33 |
-
- [ ] (待定) 根据用户找到的量化模型,更新 `app.py` 中的模型路径。
|
| 34 |
- [ ] (已暂停) 搜索 `huggingface_hub` 文档,确认是否存在用于重启 Space 的 API。
|
| 35 |
|
| 36 |
## 已完成
|
|
|
|
| 37 |
- [x] **(用户决策)** 确认 `Ling-flash-2.0` 模型过大,暂时移除,仅使用 `Ring-mini-2.0`。
|
| 38 |
- [x] 搭建 LangGraph 基础架构并重构 `app.py`。
|
| 39 |
- [x] 实现基于用户输入的模型路由逻辑。
|
|
@@ -65,4 +64,4 @@
|
|
| 65 |
- **平台:** HuggingFace Spaces
|
| 66 |
- **订阅:** HuggingFace Pro
|
| 67 |
- **推理资源:** 可以使用 ZeroGPU
|
| 68 |
-
- **文档参考:** 在必要的时候,主动搜索 HuggingFace 以及 Gradio 的在线 API 文档。
|
|
|
|
| 19 |
|
| 20 |
# 子目标
|
| 21 |
## 未完成
|
| 22 |
+
- [ ] **(进行中)** 实现自动化部署和验证流程。
|
|
|
|
| 23 |
|
| 24 |
## 已完成
|
| 25 |
+
- [x] 解决模型体积过大导致部署失败的问题。
|
| 26 |
- [x] 使用 LangGraph 实现一个可以路由两个模型的聊天网页应用。
|
| 27 |
|
| 28 |
---
|
| 29 |
|
| 30 |
# Todolist
|
| 31 |
## 未完成
|
|
|
|
|
|
|
| 32 |
- [ ] (已暂停) 搜索 `huggingface_hub` 文档,确认是否存在用于重启 Space 的 API。
|
| 33 |
|
| 34 |
## 已完成
|
| 35 |
+
- [x] 修改 `app.py`,移除 `Ling-flash-2.0` 模型,只保留 `Ring-mini-2.0`。
|
| 36 |
- [x] **(用户决策)** 确认 `Ling-flash-2.0` 模型过大,暂时移除,仅使用 `Ring-mini-2.0`。
|
| 37 |
- [x] 搭建 LangGraph 基础架构并重构 `app.py`。
|
| 38 |
- [x] 实现基于用户输入的模型路由逻辑。
|
|
|
|
| 64 |
- **平台:** HuggingFace Spaces
|
| 65 |
- **订阅:** HuggingFace Pro
|
| 66 |
- **推理资源:** 可以使用 ZeroGPU
|
| 67 |
+
- **文档参考:** 在必要的时候,主动搜索 HuggingFace 以及 Gradio 的在线 API 文档。
|
app.py
CHANGED
|
@@ -1,51 +1,28 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
-
import
|
| 3 |
-
from langchain_core.messages import AIMessage, SystemMessage, HumanMessage
|
| 4 |
|
| 5 |
-
#
|
| 6 |
-
from graph import app
|
| 7 |
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
messages = []
|
| 14 |
-
if system_message:
|
| 15 |
-
messages.append(SystemMessage(content=system_message))
|
| 16 |
-
|
| 17 |
-
for chat_message in history:
|
| 18 |
-
if chat_message['role'] == "user":
|
| 19 |
-
messages.append(HumanMessage(content=chat_message['content']))
|
| 20 |
-
elif chat_message['role'] == "assistant":
|
| 21 |
-
messages.append(AIMessage(content=chat_message['content']))
|
| 22 |
-
|
| 23 |
-
messages.append(HumanMessage(content=message))
|
| 24 |
-
|
| 25 |
-
# 使用 invoke 方法进行一次性调用
|
| 26 |
-
inputs = {"messages": messages}
|
| 27 |
-
final_state = app.invoke(inputs)
|
| 28 |
-
|
| 29 |
-
# 从最终状态中提取最后一条消息
|
| 30 |
-
final_response = final_state["messages"][-1].content
|
| 31 |
-
|
| 32 |
-
return final_response
|
| 33 |
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
respond,
|
| 37 |
-
type="messages", # 改为 messages 类型以更好地匹配 LangChain
|
| 38 |
-
additional_inputs=[
|
| 39 |
-
gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
|
| 40 |
-
],
|
| 41 |
-
)
|
| 42 |
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
|
|
|
| 48 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
|
| 50 |
if __name__ == "__main__":
|
| 51 |
demo.launch()
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
+
from comp import generate_response
|
|
|
|
| 3 |
|
| 4 |
+
# --- Gradio UI ---
|
|
|
|
| 5 |
|
| 6 |
+
with gr.Blocks() as demo:
|
| 7 |
+
gr.Markdown("# Ling Playground")
|
| 8 |
+
chatbot = gr.Chatbot()
|
| 9 |
+
msg = gr.Textbox()
|
| 10 |
+
clear = gr.ClearButton([msg, chatbot])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
+
def user(user_message, history):
|
| 13 |
+
return "", history + [[user_message, None]]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
+
def bot(history):
|
| 16 |
+
user_message = history[-1][0]
|
| 17 |
+
history[-1][1] = ""
|
| 18 |
+
for response in generate_response(user_message, history[:-1]):
|
| 19 |
+
history[-1][1] = response
|
| 20 |
+
yield history
|
| 21 |
|
| 22 |
+
msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
|
| 23 |
+
bot, chatbot, chatbot
|
| 24 |
+
)
|
| 25 |
+
clear.click(lambda: None, None, chatbot, queue=False)
|
| 26 |
|
| 27 |
if __name__ == "__main__":
|
| 28 |
demo.launch()
|
comp.py
CHANGED
|
@@ -1,12 +1,7 @@
|
|
| 1 |
import torch
|
| 2 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 3 |
-
from
|
| 4 |
-
|
| 5 |
-
import operator
|
| 6 |
-
|
| 7 |
-
# 定义此组件操作的图状态的子集
|
| 8 |
-
class GraphState(TypedDict):
|
| 9 |
-
messages: Annotated[List[AIMessage], operator.add]
|
| 10 |
|
| 11 |
# --- 模型加载 ---
|
| 12 |
# 使用 "auto" 模式加载模型和分词器,Hugging Face Accelerate 会自动处理设备和精度
|
|
@@ -20,44 +15,49 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
| 20 |
trust_remote_code=True
|
| 21 |
)
|
| 22 |
|
| 23 |
-
|
| 24 |
-
def
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
#
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
attention_mask=inputs.attention_mask.to(model.device),
|
| 56 |
-
max_new_tokens=512, # 暂时硬编码
|
| 57 |
do_sample=True,
|
| 58 |
-
|
| 59 |
)
|
| 60 |
-
output = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
|
| 61 |
|
| 62 |
-
|
| 63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
| 3 |
+
from threading import Thread
|
| 4 |
+
import spaces
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
|
| 6 |
# --- 模型加载 ---
|
| 7 |
# 使用 "auto" 模式加载模型和分词器,Hugging Face Accelerate 会自动处理设备和精度
|
|
|
|
| 15 |
trust_remote_code=True
|
| 16 |
)
|
| 17 |
|
| 18 |
+
@spaces.GPU(duration=120)
|
| 19 |
+
def generate_response(message, history):
|
| 20 |
+
# Convert history to messages format
|
| 21 |
+
messages = [
|
| 22 |
+
{"role": "system", "content": "You are Ring, an assistant created by inclusionAI"}
|
| 23 |
+
]
|
| 24 |
+
|
| 25 |
+
# Add conversation history
|
| 26 |
+
for human, assistant in history:
|
| 27 |
+
messages.append({"role": "user", "content": human})
|
| 28 |
+
messages.append({"role": "assistant", "content": assistant})
|
| 29 |
+
|
| 30 |
+
# Add current message
|
| 31 |
+
messages.append({"role": "user", "content": message})
|
| 32 |
+
|
| 33 |
+
# Apply chat template
|
| 34 |
+
text = tokenizer.apply_chat_template(
|
| 35 |
+
messages,
|
| 36 |
+
tokenize=False,
|
| 37 |
+
add_generation_prompt=True
|
| 38 |
+
)
|
| 39 |
+
|
| 40 |
+
# Tokenize input
|
| 41 |
+
model_inputs = tokenizer([text], return_tensors="pt", return_token_type_ids=False).to(model.device)
|
| 42 |
+
|
| 43 |
+
# Generate response with streaming
|
| 44 |
+
streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True, skip_prompt=True)
|
| 45 |
+
|
| 46 |
+
generation_kwargs = dict(
|
| 47 |
+
**model_inputs,
|
| 48 |
+
max_new_tokens=8192,
|
| 49 |
+
temperature=0.7,
|
|
|
|
|
|
|
| 50 |
do_sample=True,
|
| 51 |
+
streamer=streamer,
|
| 52 |
)
|
|
|
|
| 53 |
|
| 54 |
+
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
| 55 |
+
thread.start()
|
| 56 |
+
|
| 57 |
+
# Stream the response
|
| 58 |
+
response = ""
|
| 59 |
+
for new_text in streamer:
|
| 60 |
+
response += new_text
|
| 61 |
+
yield response
|
| 62 |
+
|
| 63 |
+
thread.join()
|
graph.py
DELETED
|
@@ -1,42 +0,0 @@
|
|
| 1 |
-
import operator
|
| 2 |
-
from typing import Annotated, List
|
| 3 |
-
from typing_extensions import TypedDict
|
| 4 |
-
|
| 5 |
-
from langchain_core.messages import AnyMessage
|
| 6 |
-
from langgraph.graph import StateGraph, END
|
| 7 |
-
|
| 8 |
-
# 从我们的组件文件中导入模型补全节点
|
| 9 |
-
from comp import completion_node
|
| 10 |
-
|
| 11 |
-
# --- 图状态定义 ---
|
| 12 |
-
# 状态是我们图的内存或上下文。它是一个字典,
|
| 13 |
-
# 保存了对话过程中交换的所有消息。
|
| 14 |
-
class GraphState(TypedDict):
|
| 15 |
-
"""
|
| 16 |
-
表示我们图的状态。
|
| 17 |
-
|
| 18 |
-
Attributes:
|
| 19 |
-
messages: 一个随时间自动累积的消息列表。
|
| 20 |
-
`operator.add` 注解告诉 LangGraph 将新消息附加到此列表,
|
| 21 |
-
而不是覆盖它。这就是图如何维护对话历史(上下文)的方式。
|
| 22 |
-
"""
|
| 23 |
-
messages: Annotated[List[AnyMessage], operator.add]
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
# --- 图工作流构建 ---
|
| 27 |
-
# 使用我们定义的状态创建一个新的状态图
|
| 28 |
-
workflow = StateGraph(GraphState)
|
| 29 |
-
|
| 30 |
-
# 将补全节点添加到图中。我们将其命名为 “llm”。
|
| 31 |
-
# 这个节点负责调用语言模型。
|
| 32 |
-
workflow.add_node("llm", completion_node)
|
| 33 |
-
|
| 34 |
-
# 设置图的入口点。第一个被调用的节点是 “llm”。
|
| 35 |
-
workflow.set_entry_point("llm")
|
| 36 |
-
|
| 37 |
-
# 从 “llm” 节点到 END 添加一条简单的边。
|
| 38 |
-
# 这意味着在调用 LLM 后,图的执行就完成了。
|
| 39 |
-
workflow.add_edge("llm", END)
|
| 40 |
-
|
| 41 |
-
# 将工作流编译成一个可运行的应用。
|
| 42 |
-
app = workflow.compile()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|