Qwen2.5-0.5B-Instruct

Model Overview

Qwen2.5-0.5B-Instruct is a causal language model from the Qwen2.5 series, optimized for chat and instruction-following tasks. It is available as a hosted, OpenAI-compatible API on gigarouter.

Key Capabilities

• Improved coding and mathematics performance relative to Qwen2.
• Strong instruction following, with resilience to diverse system prompts for role-play and condition-setting.
• Proficient in generating long texts (up to 8,192 tokens) and understanding structured data such as tables.
• Supports structured output generation, particularly JSON.
• Full context length of 32,768 tokens, with generation context up to 8,192 tokens.
• Multilingual support for over 29 languages including Chinese, English, French, Spanish, Portuguese, German, Italian, Russian, Japanese, Korean, Vietnamese, Thai, and Arabic.

Architecture

• Parameters: 0.49B total (0.36B non-embedding).
• Layers: 24.
• Attention heads: 14 for Q, 2 for KV (Grouped Query Attention).
• Activation: SwiGLU; normalization: RMSNorm; includes Attention QKV bias and tied word embeddings.
• Training: pretraining followed by post-training (instruction tuning).

Evaluation

Detailed evaluation results and speed benchmarks are available in the official blog and documentation.

Best For

This model is suited for applications requiring efficient, lightweight instruction-tuned chat, structured data handling, and multilingual support within a constrained parameter budget. It is particularly effective for tasks that benefit from long context or structured output generation.