minsu0567/Uni-IAD-R2-Qwen3.5-si-answer-last

VISIONConcurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 10, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

The minsu0567/Uni-IAD-R2-Qwen3.5-si-answer-last model is a fine-tuned version of the Qwen3.5-4B architecture, specifically adapted from unsloth/Qwen3.5-4B. This model has been fine-tuned on the PA_SFT_2_reordered_si_answer_last dataset, suggesting an optimization for specific instruction-following or question-answering tasks where the answer is expected at the end. Its 4 billion parameters indicate a balance between performance and computational efficiency, making it suitable for applications requiring a moderately sized yet capable language model.

Loading preview...

Model Overview

The minsu0567/Uni-IAD-R2-Qwen3.5-si-answer-last is a specialized language model derived from the Qwen3.5-4B architecture, specifically fine-tuned from the unsloth/Qwen3.5-4B base model. This fine-tuning process utilized the PA_SFT_2_reordered_si_answer_last dataset, indicating a focus on instruction-following or question-answering scenarios where the desired output format places the answer at the end of the response.

Key Characteristics

  • Base Model: Qwen3.5-4B, a 4-billion parameter model known for its general language understanding capabilities.
  • Fine-tuning Objective: Optimized for tasks where the answer is expected to be the concluding part of the generated text, as suggested by the dataset name _si_answer_last.
  • Training Configuration: The model was trained with a learning rate of 1e-05, a batch size of 1 (with 2 gradient accumulation steps), and utilized the ADAMW_BNB optimizer. Training was conducted for 1 epoch with a cosine learning rate scheduler and 100 warmup steps.

Potential Use Cases

  • Instruction Following: Generating responses to instructions where a concise, final answer is preferred.
  • Question Answering: Suitable for QA systems designed to extract and present answers at the end of a generated explanation or summary.
  • Specific Data Formats: Applications requiring outputs structured with the key information or answer positioned last, aligning with the training data's reordered format.