metadata
base_model:
- mkd-hossain/keural-alpha-base
this is finetuned model 6k
Keural-Alpha UA-SFT (6K)
This repository contains a Supervised Fine-Tuned (SFT) checkpoint of Keural-Alpha, trained to improve instruction-following and chat-style responses.
🔹 Model Overview
- Base model: keural-alpha-base
- Fine-tuning type: Supervised Fine-Tuning (Instruction / UA-style)
- Training steps: 6,000
- Context length: 2048 tokens
- Precision: bfloat16
- Framework: PyTorch + Hugging Face Transformers
- Checkpoint format:
model.safetensors
🔹 Training Data
The model was fine-tuned on cleaned instruction–response data focused on:
- Question answering
- Short explanations
- Basic reasoning tasks
- General conversational instructions
This checkpoint represents an intermediate training stage and is intended for further experimentation or continued fine-tuning.
🔹 Intended Use
- Instruction-following research
- Chat-style text generation
- Continued fine-tuning (SFT / chat datasets)
- Local and offline inference
⚠️ This is not a final model. Output quality will improve with additional training steps and more diverse chat data.