FlyPig23/Qwen3-4B_Paper_Impact_code_SFT_1ep

TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Apr 7, 2026License:otherArchitecture:Transformer Cold

FlyPig23/Qwen3-4B_Paper_Impact_code_SFT_1ep is a 4 billion parameter Qwen3-based instruction-tuned language model, fine-tuned from Qwen/Qwen3-4B-Instruct-2507. This model is specifically fine-tuned on the paper_impact_code_train dataset, indicating an optimization for tasks related to code generation or analysis within the context of research papers. It features a 32768-token context length and is designed for specialized code-related applications.

Loading preview...

Model Overview

FlyPig23/Qwen3-4B_Paper_Impact_code_SFT_1ep is a 4 billion parameter language model built upon the Qwen3 architecture, specifically fine-tuned from the Qwen/Qwen3-4B-Instruct-2507 base model. It is distinguished by its specialized training on the paper_impact_code_train dataset, suggesting a focus on tasks involving code within academic or research contexts.

Key Characteristics

  • Base Model: Qwen3-4B-Instruct-2507, a 4 billion parameter instruction-tuned model.
  • Specialized Fine-tuning: Trained for 1 epoch on the paper_impact_code_train dataset, achieving a reported evaluation loss of 0.0773.
  • Context Length: Supports a substantial context window of 32768 tokens.

Training Details

The model was trained using the following hyperparameters:

  • Learning Rate: 2e-05
  • Batch Size: 8 (train), 8 (eval) with 2 gradient accumulation steps, resulting in a total train batch size of 64.
  • Optimizer: AdamW with default betas and epsilon.
  • Scheduler: Cosine learning rate scheduler with a 0.1 warmup ratio.

Potential Use Cases

Given its fine-tuning on a code-related dataset, this model is likely suitable for:

  • Code Generation: Generating code snippets or functions based on specific requirements.
  • Code Analysis: Assisting in understanding, summarizing, or refactoring code found in research papers.
  • Research-focused Code Tasks: Applications requiring an understanding of code in an academic or scientific context.