mkd-dev's picture
Duplicate from mkd-ai-dev/keural-alpha-chat-v0.2
e7dad77
|
Raw History Blame Contribute Delete
1.27 kB
---
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.
---