Sign-SPIN Gemma2-2B (tau=0.1)

CLL Sign-SPIN (no PairRM). Frozen SFT scorer. S(y')=mean(log p + H) with tau=0.1, clip_bound=5.0, score_agg=mean.

Init: local 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, alpha=1.0, n_epochs=2, warmup_ratio=0.1, max_length=1024, prompt 512, 4 iters (ite0..ite3).

Status: ite0, ite1, ite2, and ite3 are finished. Load each finished checkpoint from iteN/, not from a LATEST/ snapshot.

from transformers import AutoModelForCausalLM
m = AutoModelForCausalLM.from_pretrained(
    "HoangTran223/SignSPIN_Gamma2-2B_0.1",
    subfolder="ite3",
)

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/SignSPIN_Gamma2-2B_0.1

Base model

google/gemma-2-2b

Finetuned

(607)

this model