Qwen: QwQ 32B Preview

QwQ-32B-Preview is an experimental AI model developed by Alibaba's Qwen team. It features 32.5 billion parameters, focusing on enhancing reasoning abilities. This model excels in math and coding tasks, achieving high scores on various benchmarks. However, it also has limitations, such as language mixing and recursive reasoning loops. Users should exercise caution when deploying it due to safety concerns.

import OpenAI from "openai"

const openai = new OpenAI({
  baseURL: "https://api.aiapilab.com/v1",
  apiKey: $AIAPILAB_API_KEY
})

async function main() {
  const completion = await openai.chat.completions.create({
    model: "qwen/qwq-32b-preview",
    messages: [
      {
        "role": "user",
        "content": "Write a blog about cat."
      }
    ]
  })

  console.log(completion.choices[0].message)
}
main()

from openai import OpenAI

client = OpenAI(
  base_url="https://api.aiapilab.com/v1",
  api_key="$AIAPILAB_API_KEY",
)

completion = client.chat.completions.create(
  model="qwen/qwq-32b-preview",
  messages=[
    {
      "role": "user",
      "content": "Write a blog about cat."
    }
  ]
)
print(completion.choices[0].message.content)

Qwen: QwQ 32B Preview

Context32768

Input$0.15 / M

Output$0.6 / M

Try Qwen: QwQ 32B Preview

Let's chat with Qwen: QwQ 32B Preview now and verify the model's response effectiveness to your questions.

What can I do for you？

Description

QwQ-32B-Preview is a new AI model from Alibaba’s Qwen team. It was recently released and has 32.5 billion parameters. This model is great at reasoning tasks. It shows strong skills in math and programming.

In benchmark tests, it scored 65% on the GP QA. It achieved 50% on AIME and over 90% on Math-500. QwQ-32B-Preview can handle inputs of up to 32,000 words. This is much longer than many other AI models can manage.

The model uses a smart way to solve problems. It plans, checks facts, and avoids common mistakes. However, it has some limits. Sometimes, it mixes languages or gets stuck in reasoning loops.

QwQ-32B-Preview has a transformer architecture. It features rotary position embeddings and attention mechanisms. The model has a context length of 32,768 tokens and includes 64 layers. It works well in logical reasoning tasks important for engineering, data science, and education.

As an open-source model under the Apache 2.0 license, it supports community collaboration. Researchers can try out its features and help improve it. You can access it on platforms like Hugging Face for testing and development.

In summary, QwQ-32B-Preview is a big step forward in reasoning AI. By using its abilities, users can solve tough problems in many areas. To get better integration options and lower prices, use our AIAPILAB services.

Model API Use Case

The QwQ-32B-Preview API is a powerful tool. It offers advanced reasoning skills for many uses. With 32.5 billion parameters, it shines in math and coding tasks. For example, it scored 90.6% on the Math-500 benchmark and 65% on the GP QA evaluation. This shows its strong analytical skills.

This API works great for educational platforms. It helps create personalized tutoring systems for students. Developers can also use it in coding environments. It improves code generation and debugging.

Researchers can use this API for scientific studies. It can handle prompts of up to 32,000 words. Its self-checking feature boosts accuracy. This reduces common errors seen in other AI models.

However, users should know it has some limits. Sometimes, it mixes languages or gets stuck in reasoning loops. Feedback from users will help make it better. For more details, check the [Hugging Face page](https://huggingface.co/Qwen/QwQ-32B-Preview).

Model Review

Pros

1. QwQ-32B-Preview demonstrates impressive reasoning skills in math and coding tasks. 2. The model processes long inputs, handling up to 32,000 words efficiently. 3. Its self-checking feature enhances accuracy and reduces common mistakes. 4. Open-source access fosters collaboration, inviting developers to improve the model. 5. QwQ-32B-Preview excels in logical reasoning, making it suitable for technical fields.

Cons

1. QwQ-32B-Preview sometimes mixes languages, causing confusion in responses. 2. The model can spiral into recursive reasoning loops, leading to lengthy, unclear answers. 3. It struggles with common sense reasoning, limiting its effectiveness in everyday tasks.

Comparison

Feature/Aspect	OpenAI GPT-4	Anthropic Claude	Qwen: QwQ-32B-Preview
Parameters	Up to 175 billion	52 billion	32.5 billion
Context Length	8,192 tokens	9,000 tokens	32,768 tokens
Language Mixing	Generally consistent language use	Consistent language use	Occasional unexpected language switching
Benchmark Performance	Strong performance across various domains	Competitive in instruction-following tasks	Excels in math and coding tasks
Reasoning Capabilities	Strong reasoning but less focus on self-checking	Emphasis on safety and ethical reasoning	Advanced reasoning with self-checking

API

import OpenAI from "openai"

const openai = new OpenAI({
  baseURL: "https://api.aiapilab.com/v1",
  apiKey: $AIAPILAB_API_KEY
})

async function main() {
  const completion = await openai.chat.completions.create({
    model: "qwen/qwq-32b-preview",
    messages: [
      {
        "role": "user",
        "content": "Write a blog about cat."
      }
    ]
  })

  console.log(completion.choices[0].message)
}
main()

from openai import OpenAI

client = OpenAI(
  base_url="https://api.aiapilab.com/v1",
  api_key="$AIAPILAB_API_KEY",
)

completion = client.chat.completions.create(
  model="qwen/qwq-32b-preview",
  messages=[
    {
      "role": "user",
      "content": "Write a blog about cat."
    }
  ]
)
print(completion.choices[0].message.content)

FAQ

Q1：What is QwQ-32B-Preview?
A1：QwQ-32B-Preview is a reasoning AI model with 32.5 billion parameters.

Q2：How can I use the QwQ-32B-Preview API?
A2：Access the API via Hugging Face; follow provided documentation for integration.

Q3：What are the main strengths of QwQ-32B-Preview?
A3：The model excels in math, coding, and logical reasoning tasks.

Q4：What limitations should I be aware of?
A4：It may mix languages, enter reasoning loops, and struggle with common sense.

Q5：How does QwQ-32B-Preview ensure response accuracy?
A5：The model fact-checks its answers and engages in self-reflection during reasoning.