lihaoxin2020/qwen3-4b-refiner-gpt54-ep2

TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Apr 18, 2026License:otherArchitecture:Transformer Cold

The lihaoxin2020/qwen3-4b-refiner-gpt54-ep2 model is a 4 billion parameter language model, fine-tuned from Qwen/Qwen3-4B-Instruct-2507 on the refiner_gpt54_sft dataset. This model is specifically adapted for tasks related to refining outputs, leveraging its base Qwen3 architecture and a 32768 token context length. It is optimized for applications requiring enhanced instruction following and refined text generation based on the specialized fine-tuning data.

Loading preview...

Model Overview

This model, lihaoxin2020/qwen3-4b-refiner-gpt54-ep2, is a 4 billion parameter language model derived from the Qwen3-4B-Instruct-2507 base architecture. It has been specifically fine-tuned on the refiner_gpt54_sft dataset, indicating an optimization for tasks involving refinement and instruction-following.

Key Characteristics

  • Base Model: Qwen/Qwen3-4B-Instruct-2507
  • Parameter Count: 4 billion parameters
  • Context Length: Supports a substantial context window of 32768 tokens.
  • Fine-tuning Focus: Specialized fine-tuning on the refiner_gpt54_sft dataset suggests enhanced capabilities in refining generated text or following complex instructions.

Training Details

The model was trained using the following hyperparameters:

  • Learning Rate: 5e-06
  • Batch Size: A total_train_batch_size of 32 (with gradient_accumulation_steps of 8 and train_batch_size of 2).
  • Optimizer: ADAMW_TORCH_FUSED with standard betas and epsilon.
  • Scheduler: Cosine learning rate scheduler with 0.05 warmup steps.
  • Epochs: Trained for 2.0 epochs.

Intended Use Cases

Given its fine-tuning on a "refiner" dataset, this model is likely suitable for applications requiring:

  • Text Refinement: Improving the quality, coherence, or style of existing text.
  • Instruction Following: Generating outputs that adhere closely to specific, detailed instructions.
  • Specialized Generation: Tasks where a refined output based on a given prompt is critical.