1. 开放权重与访问

QwQ-32B模型在Hugging Face和ModelScope上以开放权重形式发布，遵循Apache 2.0许可证，并可通过Qwen Chat访问。

2. 性能评估

QwQ-32B在一系列基准测试中进行了评估，重点关注数学推理、编码能力和一般问题解决能力。以下是QwQ-32B与其他领先模型的性能比较：

3. 使用QwQ-32B

以下是通过Hugging Face Transformers和阿里云DashScope API使用QwQ-32B的示例代码：

3.1 使用Hugging Face Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Qwen/QwQ-32B"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "How many r's are in the word \"strawberry\""
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)

model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)

运行python api_qwq.py完成测试

3.2 可视化

以下是一个使用Streamlit构建的QwQ-32B聊天机器人的示例代码，默认提供是非流式的对话平台：

# 导入所需的库
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import streamlit as st

# 在侧边栏中创建一个标题和一个链接
with st.sidebar:
    st.markdown("## Qwen/QwQ-32B LLM")
    max_length = st.slider("max_length", 0, 32768, 1024, step=1)

# 创建一个标题和一个副标题
st.title("💬 Qwen/QwQ-32B Chatbot")
st.caption("🚀 A streamlit chatbot powered by Hugging Face Transformers")

# 定义模型路径
model_name = "Qwen/QwQ-32B"

# 定义一个函数，用于获取模型和tokenizer
@st.cache_resource
def get_model():
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(
        model_name,
        torch_dtype="auto",
        device_map="auto"
    )
    model.eval()  
    return tokenizer, model

# 加载模型和tokenizer
tokenizer, model = get_model()

if "messages" not in st.session_state:
    st.session_state["messages"] = [{"role": "assistant", "content": "有什么可以帮您的？"}]

# 显示聊天记录
for msg in st.session_state.messages:
    st.chat_message(msg["role"]).write(msg["content"])

# 处理用户输入
if prompt := st.chat_input():
    st.session_state.messages.append({"role": "user", "content": prompt})
    st.chat_message("user").write(prompt)
    
    messages = [{"role": "user", "content": prompt}]
    text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    
    model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

    generated_ids = model.generate(
        **model_inputs,
        max_new_tokens=max_length
    )
    
    generated_ids = [
        output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
    ]

    response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
    st.session_state.messages.append({"role": "assistant", "content": response})
    st.chat_message("assistant").write(response)

如果想要流式的对话平台，请替换为下面的代码：

from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
import torch
import streamlit as st
from threading import Thread

# 在侧边栏中创建一个标题和一个链接
with st.sidebar:
    st.markdown("## Qwen/QwQ-32B LLM")
    max_length = st.slider("max_length", 0, 32768, 4096, step=1)
# 创建一个标题和一个副标题
st.title("💬 Qwen/QwQ-32B Chatbot")
st.caption("🚀 A streamlit chatbot powered by Hugging Face Transformers")

# 定义模型路径
model_name = "Qwen/QwQ-32B"


# 定义一个函数，用于获取模型和tokenizer
@st.cache_resource
def get_model():
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(
        model_name,
        torch_dtype="auto",
        device_map="auto"
    )
    model.eval()  
    return tokenizer, model

# 加载模型和tokenizer
tokenizer, model = get_model()

if "messages" not in st.session_state:
    st.session_state.messages = [{"role": "assistant", "content": "有什么可以帮您的？"}]

# 显示聊天记录
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"])

# 定义生成响应的函数
def generate_response(prompt):
    messages = [{"role": "user", "content": prompt}]
    text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    
    model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True)

    generation_kwargs = {
        "input_ids": model_inputs['input_ids'].to(model.device),
        "streamer": streamer,
        "max_new_tokens": max_length,  # 可以根据需要调整
        "temperature": 0.95,
        "top_p": 0.8,
    }

    # 启动生成线程
    thread = Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()
    
    generated_text = ""
    for new_text in streamer:
        generated_text += new_text
        yield generated_text  # 流式输出生成的文本

    thread.join()  # 等待线程完成

# 处理用户输入
if prompt := st.chat_input("请输入问题"):
    st.session_state.messages.append({"role": "user", "content": prompt})
    with st.chat_message("user"):
        st.markdown(prompt)

    with st.chat_message("assistant"):
        message_placeholder = st.empty()  # 创建一个占位符
        for new_text in generate_response(prompt):
            message_placeholder.markdown(new_text)  # 更新占位符内容

    # 将生成的完整内容添加到消息记录中
    st.session_state.messages.append({"role": "assistant", "content": new_text})

# 清空按钮
if st.button("清空"):
    st.session_state.messages = [{"role": "assistant", "content": "有什么可以帮您的？"}]
    st.experimental_rerun()

在终端中运行以下命令启动 streamlit 服务，并按照 compshare 的指示将端口映射到本地，然后在浏览器中打开链接 http://外部链接:11434/，即可看到聊天界面。

streamlit run /workspace/streamlit_qwq.py --server.address 0.0.0.0 --server.port 11434