Qwen3 30B A3B AWQ

🚀 Qwen3-30B-A3B-AWQ

This project provides a quantized version of the Qwen3-30B-A3B model, enabling efficient deployment and inference. It offers seamless switching between thinking and non-thinking modes, enhancing performance across various scenarios.

🚀 Quick Start

Load the Model

import torch
from modelscope import AutoModelForCausalLM, AutoTokenizer

model_name = "swift/Qwen3-30B-A3B-AWQ"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto"
)

Prepare Input and Generate Output

# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

# parsing thinking content
try:
    # rindex finding 151668 (</think>)
    index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
    index = 0

thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")

print("thinking content:", thinking_content)
print("content:", content)

✨ Features

Qwen3 Highlights

• Unique Mode Switching: Supports seamless switching between thinking and non-thinking modes within a single model, ensuring optimal performance in various scenarios.
• Enhanced Reasoning: Significantly improves reasoning capabilities, surpassing previous models in mathematics, code generation, and commonsense logical reasoning.
• Superior Alignment: Excels in human preference alignment, providing a more natural, engaging, and immersive conversational experience.
• Expert Agent Capabilities: Enables precise integration with external tools, achieving leading performance in complex agent-based tasks among open-source models.
• Multilingual Support: Supports over 100 languages and dialects, with strong capabilities for multilingual instruction following and translation.

Model Features

📦 Installation

It is recommended to use the latest version of transformers to ensure compatibility with Qwen3-MoE.

pip install transformers

💻 Usage Examples

Basic Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Qwen/Qwen3-30B-A3B"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

# parsing thinking content
try:
    # rindex finding 151668 (</think>)
    index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
    index = 0

thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")

print("thinking content:", thinking_content)
print("content:", content)

Deployment

You can use sglang>=0.4.6.post1 or vllm>=0.8.5 to create an OpenAI-compatible API endpoint.

• SGLang:

python -m sglang.launch_server --model-path Qwen/Qwen3-30B-A3B --reasoning-parser qwen3

• vLLM:

vllm serve Qwen/Qwen3-30B-A3B --enable-reasoning --reasoning-parser deepseek_r1

Local Use

Applications such as Ollama, LMStudio, MLX-LM, llama.cpp, and KTransformers have also supported Qwen3.

📚 Documentation

Switching Between Thinking and Non-Thinking Mode

• enable_thinking=True: By default, Qwen3 has thinking capabilities enabled. The model will generate think content wrapped in a <think>...</think> block, followed by the final response.

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True  # True is the default value for enable_thinking
)

⚠️ Important Note

For thinking mode, use Temperature=0.6, TopP=0.95, TopK=20, and MinP=0 (the default setting in generation_config.json). DO NOT use greedy decoding, as it can lead to performance degradation and endless repetitions.

• enable_thinking=False: This mode strictly disables the model's thinking behavior, similar to the previous Qwen2.5-Instruct models. The model will not generate any think content.

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=False  # Setting enable_thinking=False disables thinking mode
)

⚠️ Important Note

For non-thinking mode, we suggest using Temperature=0.7, TopP=0.8, TopK=20, and MinP=0.

• Advanced Usage: Switching via User Input: You can add /think and /no_think to user prompts or system messages to dynamically control the model's thinking mode.

from transformers import AutoModelForCausalLM, AutoTokenizer

class QwenChatbot:
    def __init__(self, model_name="Qwen/Qwen3-30B-A3B"):
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModelForCausalLM.from_pretrained(model_name)
        self.history = []

    def generate_response(self, user_input):
        messages = self.history + [{"role": "user", "content": user_input}]

        text = self.tokenizer.apply_chat_template(
            messages,
            tokenize=False,
            add_generation_prompt=True
        )

        inputs = self.tokenizer(text, return_tensors="pt")
        response_ids = self.model.generate(**inputs, max_new_tokens=32768)[0][len(inputs.input_ids[0]):].tolist()
        response = self.tokenizer.decode(response_ids, skip_special_tokens=True)

        # Update history
        self.history.append({"role": "user", "content": user_input})
        self.history.append({"role": "assistant", "content": response})

        return response

# Example Usage
if __name__ == "__main__":
    chatbot = QwenChatbot()

    # First input (without /think or /no_think tags, thinking mode is enabled by default)
    user_input_1 = "How many r's in strawberries?"
    print(f"User: {user_input_1}")
    response_1 = chatbot.generate_response(user_input_1)
    print(f"Bot: {response_1}")
    print("----------------------")

    # Second input with /no_think
    user_input_2 = "Then, how many r's in blueberries? /no_think"
    print(f"User: {user_input_2}")
    response_2 = chatbot.generate_response(user_input_2)
    print(f"Bot: {response_2}") 
    print("----------------------")

    # Third input with /think
    user_input_3 = "Really? /think"
    print(f"User: {user_input_3}")
    response_3 = chatbot.generate_response(user_input_3)
    print(f"Bot: {response_3}")

⚠️ Important Note

When enable_thinking=True, the model will always output a <think>...</think> block, but the content may be empty. When enable_thinking=False, the soft switches are not valid.

Agentic Use

Qwen3 excels in tool calling capabilities. We recommend using Qwen-Agent to simplify the process.

from qwen_agent.agents import Assistant

# Define LLM
llm_cfg = {
    'model': 'Qwen3-30B-A3B',

    # Use the endpoint provided by Alibaba Model Studio:
    # 'model_type': 'qwen_dashscope',
    # 'api_key': os.getenv('DASHSCOPE_API_KEY'),

    # Use a custom endpoint compatible with OpenAI API:
    'model_server': 'http://localhost:8000/v1',  # api_base
    'api_key': 'EMPTY',

    # Other parameters:
    # 'generate_cfg': {
    #         # Add: When the response content is `<think>this is the thought</think>this is the answer;
    #         # Do not add: When the response has been separated by reasoning_content and content.
    #         'thought_in_content': True,
    #     },
}

# Define Tools
tools = [
    {'mcpServers': {  # You can specify the MCP configuration file
            'time': {
                'command': 'uvx',
                'args': ['mcp-server-time', '--local-timezone=Asia/Shanghai']
            },
            "fetch": {
                "command": "uvx",
                "args": ["mcp-server-fetch"]
            }
        }
    },
  'code_interpreter',  # Built-in tools
]

# Define Agent
bot = Assistant(llm=llm_cfg, function_list=tools)

# Streaming generation
messages = [{'role': 'user', 'content': 'https://qwenlm.github.io/blog/ Introduce the latest developments of Qwen'}]
for responses in bot.run(messages=messages):
    pass
print(responses)

Processing Long Texts

Qwen3 natively supports a context length of up to 32,768 tokens. For longer texts, you can use YaRN scaling techniques.

• Modify Model Files: Add the rope_scaling fields to the config.json file.

{
    ...,
    "rope_scaling": {
        "rope_type": "yarn",
        "factor": 4.0,
        "original_max_position_embeddings": 32768
    }
}

• Pass Command Line Arguments:
  • vLLM:

vllm serve ... --rope-scaling '{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}' --max-model-len 131072

• sglang:

python -m sglang.launch_server ... --json-model-override-args '{"rope_scaling":{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}}'

• llama-server:

llama-server ... --rope-scaling yarn --rope-scale 4 --yarn-orig-ctx 32768

⚠️ Important Note

If you encounter the warning Unrecognized keys in rope_scaling for 'rope_type'='yarn': {'original_max_position_embeddings'}, please upgrade transformers>=4.51.0.

💡 Usage Tip

The endpoint provided by Alibaba Model Studio supports dynamic YaRN by default, no extra configuration is needed.

🔧 Technical Details

Best Practices

1. Sampling Parameters:
   • Thinking mode: Temperature=0.6, TopP=0.95, TopK=20, MinP=0.
   • Non-thinking mode: Temperature=0.7, TopP=0.8, TopK=20, MinP=0.
   • Adjust presence_penalty between 0 and 2 to reduce repetitions.
2. Output Length: Use 32,768 tokens for most queries, and 38,912 tokens for complex problems.
3. Standardize Output Format:
   • Math problems: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
   • Multiple-choice questions: Add "Please show your choice in the answer field with only the choice letter, e.g., "answer": "C"." to the prompt.
4. No Thinking Content in History: In multi-turn conversations, only include the final output in the history.

Citation

@misc{qwen3,
    title  = {Qwen3},
    url    = {https://qwenlm.github.io/blog/qwen3/},
    author = {Qwen Team},
    month  = {April},
    year   = {2025}
}

📄 License

This project is licensed under the Apache-2.0 license.