Threads Qwen3-8B LoRA
QLoRA fine-tuned adapter for generating Threads-style English tech-news posts.
Model Details
- Base model: Qwen/Qwen3-8B
- Method: QLoRA (4-bit NF4, rank 8, alpha 16)
- Epochs: 2
- Learning rate: 0.0001
- Max sequence length: 1024
Training Data
Fine-tuned on the theartificialintelligence subset of
e1879/threads-english-tech-news-sft,
a supervised-pairs dataset of Threads tech-news posts with reverse-engineered instructions.
Usage
from peft import AutoPeftModelForCausalLM
from transformers import AutoTokenizer
model = AutoPeftModelForCausalLM.from_pretrained(
"e1879/threads-qwen3-8b-lora",
load_in_4bit=True,
device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("e1879/threads-qwen3-8b-lora")
- Downloads last month
- 1