Qwen's QwQ-32B: Small Model with Huge Potential - Analytics Vidhya

China's AI prowess is rapidly expanding, with models like DeepSeek and Qwen challenging global leaders. DeepSeek, a ChatGPT rival, has garnered significant attention, while Qwen's versatile chatbot, integrating vision, reasoning, and coding, is making impressive strides. QwQ 32B, Qwen's latest reasoning model, is a mid-sized contender, competing with top-tier models like DeepSeek-R1 and o1-mini, demonstrating China's remarkable advancements in AI.

Table of Contents

• Understanding Qwen's QwQ 32B
• Performance Benchmarks
• Accessing QwQ 32B:
  • The Easiest Method: Qwen Chat
  • Local Deployment via Hugging Face
  • Simplified Local Setup with Ollama
• QwQ 32B in Action
• Conclusion

Understanding Qwen's QwQ 32B

QwQ-32B, a 32-billion parameter model from the Qwen family, leverages Reinforcement Learning (RL) to enhance its reasoning and problem-solving capabilities. Its performance rivals that of larger models such as DeepSeek-R1, adapting its reasoning based on feedback and effectively utilizing tools. Open-weight and available under the Apache 2.0 license on Hugging Face and ModelScope, it's also accessible through Qwen Chat, showcasing RL's potential to significantly boost AI performance.

Performance Benchmarks

QwQ-32B's mathematical reasoning, coding, and problem-solving skills have been rigorously tested across various benchmarks. The following comparisons highlight its performance against leading models like DeepSeek-R1-Distilled-Qwen-32B, DeepSeek-R1-Distilled-Llama-70B, o1-mini, and the original DeepSeek-R1.

LiveBench scores, evaluating reasoning across diverse tasks, position QwQ-32B between R1 and o3-mini, yet at a significantly lower cost (approximately 1/10th). Pricing estimates, based on API or OpenRouter data, place QwQ-Preview at $0.18 per output token on DeepInfra, emphasizing its cost-effectiveness.

Alibaba's QwQ-32B achieves a 59% score on GPQA Diamond (scientific reasoning) and 86% on AIME 2024 (mathematics). While excelling in math, its scientific reasoning lags behind top competitors.

Currently trending #1 on HuggingFace.

Learn more through our free QwQ 32B course!

Accessing QwQ 32B

Accessing QwQ-32B offers several options depending on your needs and technical expertise.

Via Qwen Chat (Simplest Approach)

1. Visit https://www.php.cn/link/e3524b4d458e3625befde27f60809f34.
2. Create an account (if needed).
3. Select "QwQ-32B" from the model selection menu.
4. Begin interacting with the model.

Local Deployment via Hugging Face

Prerequisites:

• High-end GPU (24GB VRAM minimum; 80GB for unquantized FP16; around 20GB for quantized versions).
• Python 3.8 , Git, pip or conda.
• Hugging Face transformers library (4.37.0 ).

Installation and Usage: (Code snippets provided in the original text are retained here)

<code>pip install transformers torch</code>

<code>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)</code>

<code>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=512)
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)</code>

Simplified Local Setup with Ollama

1. Download and install Ollama from ollama.com.
2. Pull the model: ollama pull qwq:32b
3. Run the model: ollama run qwq:32b

QwQ 32B in Action

(Examples with embedded videos are retained from the original text)

Prompt: Create a static webpage with illuminating candle with sparks around the flame

Prompt: Develop a seated game where you can fire missiles in all directions. At first, the enemy’s speed is very slow, but after defeating three enemies, the speed gradually increases. implement in p5.js

Prompt: Write a Python program that shows a ball bouncing inside a spinning hexagon. The ball should be affected by gravity and friction, and it must bounce off the rotating walls realistically.

Conclusion

QwQ-32B represents a substantial advancement in AI reasoning, offering performance comparable to top models at a fraction of the cost. Its strong LiveBench scores and cost-effectiveness ($0.18 per output token) make it a practical and accessible solution for diverse applications. This progress signifies the potential for high-performance AI to become more affordable and widely accessible, fostering greater innovation.

Learn more about using QwQ 32B in your projects with our free course!

The above is the detailed content of Qwen's QwQ-32B: Small Model with Huge Potential - Analytics Vidhya. For more information, please follow other related articles on the PHP Chinese website!