Qwen3.5-0.8B-Instruct-Demo

Igbo fine-tune of Qwen3.5-0.8B via multi-session relay race fine-tuning.

Trained across multiple Colab T4 sessions using Unsloth + QLoRA 4-bit.

Training

Hyperparameter	Value
Base model	Qwen/Qwen3.5-0.8B-Instruct
Method	QLoRA 4-bit, r=16
Sessions	3 Colab T4 relay-race checkpoints
Checkpoint relay	push_to_hub every 1000 steps
Framework	Unsloth

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("KiriLabs/Qwen3.5-0.8B-Instruct-Demo")
tokenizer = AutoTokenizer.from_pretrained("KiriLabs/Qwen3.5-0.8B-Instruct-Demo")

Example

Input:  What is the Igbo word for 'water'?
Output: Mmiri