Qwen2.5-0.5B-Portuguese-v1
| Property | Value |
|---|---|
| Parameter Count | 494M parameters |
| Model Type | Causal Language Model |
| Memory Footprint | 988MB |
| Maximum Context Length | 480 tokens |
| Model URL | HuggingFace |
What is Qwen2.5-0.5B-Portuguese-v1?
Qwen2.5-0.5B-Portuguese-v1 is a specialized Portuguese language model based on the Qwen2.5 architecture, fine-tuned specifically for enhanced Portuguese language understanding and generation. This model represents a significant advancement in Portuguese NLP capabilities, offering a compact yet powerful solution with 494M parameters.
Implementation Details
The model utilizes the transformers library and runs on CUDA-enabled devices with automatic dtype optimization. It features a context window of 480 tokens and generates responses up to 512 tokens. The implementation includes built-in chat templating and system prompts optimized for Portuguese language tasks.
- Implements bfloat16 precision for efficient computation
- Supports batch processing with automated device mapping
- Features built-in chat template functionality
- Includes automatic system prompt injection for optimal Portuguese performance
Core Capabilities
- Strong performance on Portuguese NLI tasks (79.1% accuracy on FaQuAD-NLI)
- Effective hate speech detection (70.2% accuracy)
- Capable of handling complex academic content (ENEM and OAB exam tasks)
- Sentiment analysis capabilities (59.4% accuracy on TweetSentBR)
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its specialized Portuguese language capabilities while maintaining a relatively small parameter count of 494M, making it efficient for deployment while still achieving strong performance across various Portuguese NLP tasks.
Q: What are the recommended use cases?
The model is well-suited for Portuguese language tasks including text generation, question answering, sentiment analysis, and content classification. It performs particularly well in academic and legal contexts, as demonstrated by its performance on ENEM and OAB exam benchmarks.
