Pq234/gemma-4-12B-coder-fable5-composer2.5-v1
Pq234/gemma-4-12B-coder-fable5-composer2.5-v1 is a 12 billion parameter Gemma 4 based model fine-tuned by Pq234, specifically optimized for verifiable Python code generation and agentic reasoning. It excels at producing clean, runnable Python solutions by reasoning through edge cases and complexity. This model features an extended context length of 256K tokens, making it suitable for complex coding tasks.
Loading preview...
Overview
Pq234/gemma-4-12B-coder-fable5-composer2.5-v1 is a 12 billion parameter Gemma 4 model, fine-tuned for verifiable Python coding and agentic behavior. This full-precision safetensors master is designed for developers to create custom quantizations (GGUF, MLX, AWQ, GPTQ) or for further fine-tuning. It features a corrected and extended context length of 256K tokens.
Key Capabilities
- High-Quality Python Code Generation: Produces clean, runnable Python solutions after reasoning through problem complexities and edge cases.
- Agentic Reasoning: Significantly enhanced agentic capabilities, especially in the upcoming v2, allowing it to "think" in a native thought channel before generating code.
- Verifiable Training Data: Trained on a unique dataset combining Composer 2.5's real Chain-of-Thought (CoT) and Fable 5's re-derived CoT for hard cases, with all solutions verified by execution.
- Reduced Refusals: Task-focused training minimizes safety hedging, making it less prone to refusals than the base model.
Use Cases
- Custom Quantization: Ideal for users who need to roll their own GGUF, MLX, AWQ, or GPTQ builds.
- Further Fine-tuning: Provides a clean base for continued training or LoRA adaptations.
- Python Development: Excellent for generating Python functions, solving algorithmic problems, and handling complex coding challenges.
- Speculative Decoding: Compatible with Gemma 4 MTP draft models for lossless speculative decoding, offering faster inference.