Ridvii/qwen2.5-14b-bangla-hallucination-16bit
The Ridvii/qwen2.5-14b-bangla-hallucination-16bit is a 14.8 billion parameter Qwen2.5 model, developed by Ridvii, fine-tuned from unsloth/Qwen2.5-14B-Instruct-bnb-4bit. This model was trained using Unsloth and Huggingface's TRL library, enabling faster training. It is designed for tasks related to the Bengali language, particularly focusing on hallucination reduction within that context, and supports a 32768 token context length.
Loading preview...
Model Overview
Ridvii/qwen2.5-14b-bangla-hallucination-16bit is a 14.8 billion parameter language model developed by Ridvii. It is fine-tuned from the unsloth/Qwen2.5-14B-Instruct-bnb-4bit base model, leveraging the Qwen2.5 architecture. The training process utilized Unsloth and Huggingface's TRL library, which facilitated a 2x faster fine-tuning compared to standard methods.
Key Characteristics
- Base Model: Qwen2.5-14B-Instruct-bnb-4bit.
- Parameter Count: 14.8 billion parameters.
- Training Efficiency: Fine-tuned with Unsloth for accelerated training.
- Context Length: Supports a substantial context window of 32768 tokens.
- License: Released under the Apache-2.0 license.
Intended Use Cases
This model is particularly suited for applications requiring a large language model with a focus on the Bengali language. Its fine-tuning suggests an emphasis on addressing or reducing hallucination issues, making it potentially valuable for tasks where factual accuracy and coherent generation in Bengali are critical. Developers can integrate this model into projects that benefit from its extensive parameter count and optimized training methodology.