didula-wso2/exp_24_sft-activesft_16bit_vllm

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Mar 5, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

The didula-wso2/exp_24_sft-activesft_16bit_vllm is a 7.6 billion parameter Qwen2-based causal language model, fine-tuned from unsloth/qwen2.5-coder-7b-instruct-bnb-4bit. Developed by didula-wso2, this model was trained using Unsloth and Huggingface's TRL library for accelerated performance. It is optimized for tasks leveraging its Qwen2 architecture and instruction-following capabilities, offering a 32768 token context length.

Loading preview...

Overview

This model, developed by didula-wso2, is a 7.6 billion parameter Qwen2-based causal language model. It was fine-tuned from the unsloth/qwen2.5-coder-7b-instruct-bnb-4bit base model, indicating a focus on instruction-following and potentially coding-related tasks given its origin. The training process utilized Unsloth and Huggingface's TRL library, which enabled a 2x faster fine-tuning speed.

Key Characteristics

  • Base Model: Qwen2.5-Coder-7B-Instruct
  • Parameter Count: 7.6 billion parameters
  • Context Length: 32768 tokens
  • Training Method: Fine-tuned with Unsloth and Huggingface TRL for accelerated training.

Potential Use Cases

Given its foundation in a coder-instruct model and instruction-tuned nature, this model is likely suitable for:

  • General instruction following tasks.
  • Code generation and completion.
  • Text summarization and question answering based on provided instructions.
  • Applications requiring a balance of performance and efficiency due to its optimized training.