ermiaazarkhalili/Qwen3.5-4B-SFT-Claude-Opus-Reasoning-Unsloth

VISIONConcurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Apr 25, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The ermiaazarkhalili/Qwen3.5-4B-SFT-Claude-Opus-Reasoning-Unsloth is a 4.7 billion parameter language model, fine-tuned from unsloth/Qwen3.5-4B. This model was supervised fine-tuned using LoRA on a private Claude reasoning distillation dataset, specifically targeting improved reasoning capabilities. It is designed for instruction-following tasks, inheriting the Qwen3.5 architecture and an Apache-2.0 license.

Loading preview...

Model Overview

This model, ermiaazarkhalili/Qwen3.5-4B-SFT-Claude-Opus-Reasoning-Unsloth, is a 4.7 billion parameter language model based on the Qwen3.5ForConditionalGeneration architecture. It is a LoRA fine-tune of the unsloth/Qwen3.5-4B base model, specifically optimized through supervised fine-tuning (SFT) using the ermiaazarkhalili/claude-reasoning-distillation dataset.

Key Characteristics

  • Base Model: unsloth/Qwen3.5-4B
  • Fine-tuning Method: LoRA SFT via Unsloth and TRL libraries.
  • Training Data: Private ermiaazarkhalili/claude-reasoning-distillation dataset, configured for SFT.
  • LoRA Configuration: Utilizes a LoRA rank (r) of 64 and alpha of 64, with a learning rate of 0.0002 over 1 epoch.
  • Precision: Trained in 4-bit QLoRA precision.
  • Max Sequence Length: Supports a maximum sequence length of 2048 tokens.

Limitations

  • No Benchmarks: No downstream benchmark evaluations have been conducted; only training loss observations are available.
  • Inherited Biases: Inherits biases, knowledge cutoff, and potential failure modes from its base model.
  • Specific Fine-tuning: Its behavior outside the distribution of the single instruction-following dataset it was fine-tuned on is untested.
  • Merged Adapters: LoRA adapters are merged into the base weights, preventing detachment from this fine-tune.