Modelpublic

prometheus04/vibethinker-1.5b-rstar-coder

sourceHugging Faceapache-2.0updated 7mo agoView on Hugging Face

0likes10downloads

Model Card

VibeThinker-1.5B + rStar-Coder

Fine-tuned VibeThinker-1.5B on 80K Python samples from microsoft/rStar-Coder.

Training:

—LoRA r=16, alpha=32
—1 epoch, batch=16
—Cost: $3-4 on A100

Usage:

python

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("prometheus04/vibethinker-1.5b-rstar-coder", torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("prometheus04/vibethinker-1.5b-rstar-coder")