SPIN Gemma2-2B

SPIN (DPO self-play, no PairRM) from SFT gemma2_2b_sft_ultrachat200k_20260820_043159/final.

Locked recipe: batch=2, grad_accum=2, lr 5e-7 ite0–1 then 1e-7 ite2–3, beta=0.1, n_epochs=2, warmup_ratio=0.1, max_length=1024, prompt 512, 4 iters.

Status: ite0 and ite1 finished (weights at iteN/ root). ite2 is in progress, resume snapshot at ite2/LATEST (example_counter=58000). ite2/train.jsonl is the generated self-play set.

from transformers import AutoModelForCausalLM
m = AutoModelForCausalLM.from_pretrained("HoangTran223/SPIN_Gamma2-2B", subfolder="ite1")

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for HoangTran223/SPIN_Gamma2-2B

Base model

google/gemma-2-2b

Finetuned

(607)

this model