AmanKumarAryan/Deku

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 29, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Deku is a 7.6 billion parameter Qwen2-based instruction-tuned causal language model developed by AmanKumarAryan. This model was fine-tuned using Unsloth and Huggingface's TRL library, specifically building upon the unsloth/qwen2.5-coder-7b-instruct-bnb-4bit base. It is optimized for efficient training and aims to provide strong performance for general instruction-following tasks.

Loading preview...

Deku Model Overview

Deku is a 7.6 billion parameter instruction-tuned language model developed by AmanKumarAryan. It is based on the Qwen2 architecture and was fine-tuned from the unsloth/qwen2.5-coder-7b-instruct-bnb-4bit model. The fine-tuning process leveraged Unsloth and Huggingface's TRL library, which enabled a significantly faster training time.

Key Characteristics

  • Base Model: Qwen2.5-Coder-7B-Instruct
  • Parameter Count: 7.6 billion
  • Training Efficiency: Utilizes Unsloth for 2x faster fine-tuning.
  • License: Apache-2.0, allowing for broad use and distribution.

Intended Use Cases

Deku is suitable for a variety of instruction-following tasks, benefiting from its Qwen2.5-Coder base. While specific benchmarks are not provided, its foundation suggests potential for:

  • General conversational AI.
  • Code-related tasks, given its base model's focus.
  • Text generation and summarization.
  • Applications requiring an efficiently trained, medium-sized language model.