Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled

🌟 Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled#

🔥 Update (April 5): I’ve released the complete training notebook, codebase, and a comprehensive PDF guide to help beginners and enthusiasts understand and reproduce this model's fine-tuning process.

❤️ Special thanks to the Unsloth open-source library and @KyleHessling1 for their support.

📚 Resources & Guides#

👉 GitHub Repository: Jackrong-llm-finetuning-guide Visit the repo to dive into the codebase and reproduce the results locally or on Colab.

📥 Core Technical Document#

🔗 Qwopus3.5-27b Complete Fine-Tuning Guide (PDF)

• The Full Pipeline: A step-by-step walkthrough—from downloading the base model and unifying heterogeneous data, to configuring trainer hyperparameters and publishing to Hugging Face.
• Beginner Friendly: Includes an introductory guide to getting started with Google Colab and Unsloth.
• Feedback welcome! If you spot any areas for improvement, please let me know and I will update it promptly.

A Note: My goal isn't just to detail a workflow, but to demystify LLM training. Beyond the social media hype, fine-tuning isn't an unattainable ritual—often, all you need is a Google account, a standard laptop, and relentless curiosity.

No one starts as an expert, but every expert was once brave enough to begin.

All training and testing for this project were self-funded. If you find this model or guide helpful, a Star ⭐️ on GitHub would be the greatest encouragement. Thank you! 🙏

Note. The Claude series model optimizations are named under the Qwopus3.5 series, with the latest version being 🌟Qwopus3.5-v3.

📢 Announcement#

Update: This model has been further enhanced with additional reasoning data distilled from Qwen3.5-27B.

The new training data introduces higher-quality reasoning trajectories across domains such as science, instruction-following, and mathematics.

Part of the data comes from Jackrong/Qwen3.5-reasoning-700x, a curated dataset designed to improve structured step-by-step reasoning and reasoning diversity.

Image: HCaJnUQaoAAaMIc

💡 Model Introduction#

Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled is a highly capable reasoning model fine-tuned on top of the Qwen3.5-4B dense architecture. The model's core directive is to leverage state-of-the-art Chain-of-Thought (CoT) distillation primarily sourced from Claude-4.6 Opus interactions.

Through Supervised Fine-Tuning (SFT) focusing specifically on structured reasoning logic, this model excels in breaking down complex user problems, planning step-by-step methodologies within strictly formatted <think> tags, and ultimately delivering precise, nuanced solutions.

🗺️ Training Pipeline Overview#

Base Model (Qwen3.5-4B)
 │
 ▼
Supervised Fine-Tuning (SFT) + LoRA
(Response-Only Training masked on "<|im_start|>assistant\n<think>")
 │
 ▼
Final Model Text Only (Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled)

🧠 Example of Learned Reasoning Scaffold（Example）#

The model includes targeted optimizations addressing Qwen3.5’s tendency toward excessive transitional or repetitive reasoning on simple queries. Through deep distillation and structural imitation of Claude-4.6-Opus reasoning chains, the model adopts a more efficient structured thinking pattern: “Let me analyze this request carefully: 1..2..3...”. This streamlined reasoning paradigm significantly reduces redundant cognitive loops while preserving deep analytical capacity, resulting in substantially improved inference efficiency.

Let me analyze this request carefully:

1. Identify the core objective of the problem.
2. Break the task into clearly defined subcomponents.
3. Evaluate constraints and edge cases.
4. Formulate a step-by-step solution plan.
5. Execute the reasoning sequentially and verify consistency.
            .
            .
            .

🔹 Supervised Fine-Tuning (SFT)#

• Objective: To inject high-density reasoning logic and establish a strict format for problem-solving involving an internal thinking state prior to outputting the final response.
• Method: We utilized Unsloth for highly efficient memory and compute optimization. A critical component of this stage is the train_on_responses_only strategy, masking instructions so the loss is purely calculated over the generation of the <think> sequences and the subsequent solutions.
• Format Enforcement: All training samples were systematically normalized so the model strictly abides by the structure <think> {internal reasoning} </think>\n {final answer}.

📈 Training Loss Curve#

The training loss showed a strong and healthy downward trend throughout the run, demonstrating effective knowledge distillation. Starting from an initial loss of 0.74356, the model converged steadily to a final loss of 0.23984 — indicating the model successfully internalized the structured <think> reasoning patterns from the Claude 4.6 Opus teacher data.

📚 All Datasets Used#

The dataset consists of high-quality, filtered reasoning distillation data:

🌟 Benchmark#

These evaluation results were originally reported by khitsly. Evaluation was conducted using the EleutherAI lm-evaluation with 8-bit inference and temperature 0. Higher scores indicate better performance.

🌟 Core Skills & Capabilities#

1. Modular & Structured Thinking: Inheriting traits from Opus-level reasoning, the model demonstrates confident parsing of the prompt, establishing an outlined plan in its <think> block sequentially rather than exploratory "trial-and-error" self-doubt.
2. Extended Context Support: Fine-tuned smoothly with a 16,384 token context window allowing complex multi-step reasoning traces to exist gracefully within memory limits.

⚠️ Limitations & Intended Use#

• Hallucination Risk: While reasoning is strong, the model remains an autoregressive LLM; external facts provided during the thinking sequence may occasionally contain hallucinations if verifying real-world events.
• Intended Scenario: Best suited for offline analytical tasks, coding, math, and heavy logic-dependent prompting where the user needs to transparently follow the AI's internal logic.
• This model is a test version intended solely for learning and demonstration purposes, and is for academic research and technical exploration use only.

🙏 Acknowledgements#

Significant thanks to the Unsloth AI team for making rapid fine-tuning of large LLM models accessible. Additionally, we acknowledge Qwen internally, and the open-source community developers producing exceptional distilled datasets (nohurry and TeichAI).