ertghiu256/qwen3-4b-code-reasoning-v2

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 2, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The ertghiu256/qwen3-4b-code-reasoning-v2 is a 4 billion parameter Qwen3 model developed by ertghiu256, fine-tuned for code reasoning and instruction. It was trained on nvidia/OpenCodeReasoning and nvidia/OpenCodeInstruct datasets, with a context length up to 32,000 tokens. This model specializes in code-related tasks, leveraging its training on specific code reasoning and instruction datasets.

Loading preview...

Overview

The ertghiu256/qwen3-4b-code-reasoning-v2 is a 4 billion parameter Qwen3 model, developed by ertghiu256. It has been specifically fine-tuned for code reasoning and instruction tasks, utilizing the nvidia/OpenCodeReasoning and nvidia/OpenCodeInstruct datasets. The training data was filtered to ensure a context length of up to 32,000 tokens, making it suitable for handling moderately long code snippets and instructions.

Key Capabilities

  • Code Reasoning: Optimized for understanding and processing logical structures within code.
  • Code Instruction Following: Designed to accurately interpret and execute code-related instructions.
  • Efficient Training: Fine-tuned using Unsloth and Huggingface's TRL library, resulting in 2x faster training.
  • Anchored Knowledge: Enhanced with a subset of ChavyvAkvar/qwen3-coder-480b-distill-mini-Converted for broader code understanding.

Recommended Usage

This model is ideal for applications requiring robust code understanding and generation based on instructions. Recommended inference parameters include a temperature of 0.6-0.7, a repeat penalty of 1.2, top_p of 0.95, and top_k of 20, with enable_thinking set to True for potentially improved reasoning.