Enterprise Plans
AI Application Development
AI Translator API
AI SEO/GEO Service
Web Scraping Service
Log in
const { OpenAI } = require('openai');
const api = new OpenAI({ apiKey: '', baseURL: 'https://api.ai.cc/v1' });
const main = async () => {
const prompt = `
All of the states in the USA:
- Alabama, Mongomery;
- Arkansas, Little Rock;
`;
const response = await api.completions.create({
prompt,
model: 'Qwen/Qwen1.5-1.8B',
});
const text = response.choices[0].text;
console.log('Completion:', text);
};
main();
from openai import OpenAI
client = OpenAI(
api_key="",
base_url="https://api.ai.cc/v1",
)
def main():
response = client.completions.create(
model="Qwen/Qwen1.5-1.8B",
prompt="""
All of the states in the USA:
- Alabama, Mongomery;
- Arkansas, Little Rock;
""",
)
completion = response.choices[0].text
print(f"Completion: {completion}")
main()
-
AI Playground
Test all API models in the sandbox environment before you integrate.
We provide more than 300 models to integrate into your app.
.webp)
Product Detail
Discover the powerful Qwen 1.5 (1.8B), the latest advancement in Alibaba Cloud's cutting-edge Qwen series of large language models. This impressive series offers a diverse range of models, scaling from 0.5 billion up to a massive 72 billion parameters. Designed to outperform its rivals, Qwen 1.5 brings significant improvements in both raw performance and alignment with human preferences, setting a new standard for accessible and robust AI.
✨ Unpacking Qwen 1.5 (1.8B): Core Description
The Qwen 1.5 (1.8B) model, an early release from the Qwen2 series, is a sophisticated transformer-based, decoder-only language model. It has undergone extensive pre-training on a substantial corpus of data, ensuring broad knowledge and strong generative capabilities. The Qwen 1.5 family encompasses a variety of sizes: 0.5B, 1.8B, 4B, 7B, 14B, and 72B. Each size is available as both a base language model and an instruction-aligned chat model, catering to diverse application needs.
Its foundational architecture is built upon the advanced Transformer design, incorporating key innovations such as:
- ✅ SwiGLU Activation: Enhances model non-linearity and performance.
- ✅ Attention QKV Bias: Improves attention mechanism stability and effectiveness.
- ✅ Group Query Attention: Optimizes efficiency for larger models.
- ✅ Hybrid Attention: A combination of sliding window attention and full attention to manage context effectively.
💡 Context Length: Qwen 1.5 supports an impressive 32K token context length, allowing it to process and generate significantly longer and more complex text sequences, crucial for advanced applications.
🌐 Multilingual Capabilities: A key strength is its enhanced multilingual support, featuring an improved tokenizer specifically adapted to handle a wide array of natural languages and programming codes with high accuracy.
🚀 Qwen 1.5 Against the Competition
In the rapidly evolving landscape of large language models, Qwen 1.5 presents formidable competition. When benchmarked against industry leaders such as Claude 2.1, GPT-3.5-Turbo, and Mixtral, Qwen 1.5 consistently demonstrates superior performance across various metrics.
Across traditional benchmarks covering fundamental capabilities like language understanding and reasoning, Qwen 1.5 shows exceptional strength. Furthermore, its chat models have achieved impressive scores on widely recognized benchmarks such as MT-Bench and AlpacaEval, indicating a high degree of alignment with human preferences and conversational quality.
Illustration of Qwen 1.5's fine-tuning performance.
The model's multilingual prowess is also noteworthy, demonstrating strong results across a diverse set of languages. It has been rigorously evaluated on numerous benchmarks encompassing exams, understanding tasks, translation, and mathematical problem-solving, solidifying its versatility for global applications.
🛠️ Essential Usage Tips for Qwen 1.5
-
⚙️ Installation: To ensure smooth operation and prevent common errors, it is highly recommended to install
transformers>=4.37.0. - ⚠️ Base Model Caution: For optimal text generation quality, it is generally advisable not to use the base language models directly.
- 💡 Post-Training Techniques: To unlock the full potential of Qwen 1.5, consider applying post-training techniques such as Supervised Fine-Tuning (SFT), Reinforcement Learning from Human Feedback (RLHF), or continued pretraining on the model for specific tasks or domains.
⚖️ License Agreement for Qwen 1.5
Each model within the Qwen 1.5 series has its specific license detailed inside its Hugging Face repository. Importantly, you do NOT need to submit a special request for commercial usage, simplifying deployment for businesses and developers.
🌟 Conclusion: The Future of LLMs with Qwen 1.5
The Qwen 1.5 (1.8B) model signifies a pivotal achievement in the realm of large language models. Its impressive capabilities, coupled with competitive performance and robust multilingual support, position it as a highly promising tool for a wide array of AI applications. As this model continues to evolve, we can anticipate even more advanced features and further performance enhancements, making it a critical asset for innovation.
❓ Frequently Asked Questions (FAQs)
Q: What is Qwen 1.5 (1.8B)?
A: Qwen 1.5 (1.8B) is the beta version of Qwen2, a transformer-based, decoder-only large language model developed by Alibaba Cloud, notable for its balance of performance and efficiency.
Q: What are the key architectural features of Qwen 1.5?
A: It uses a Transformer architecture with SwiGLU activation, attention QKV bias, group query attention, and a mixture of sliding window and full attention, supporting a 32K token context length.
Q: How does Qwen 1.5 compare to other LLMs like GPT-3.5-Turbo or Mixtral?
A: Qwen 1.5 demonstrates superior performance across basic capabilities, language understanding, reasoning, and alignment with human preferences on various benchmarks.
Q: Can I use Qwen 1.5 for commercial purposes?
A: Yes, you can. You do not need to submit a special request for commercial usage. Please refer to the specific license in each model's Hugging Face repository.
Q: Are there any recommendations for using Qwen 1.5 effectively?
A: It's recommended to install transformers>=4.37.0 and to apply post-training techniques like SFT or RLHF rather than using the base models directly for text generation.
Learn how you can transformyour company with AICC APIs
