Modelpublic

Kausp11/searchlm-wiki6m-search_1p7B_tier_v2_fast

sourceHugging Faceapache-2.0updated 13d agoView on Hugging Face

0likes38downloads

Model Card

wiki6m/search1p7Btierv2fast

Role: 1.7B search arm, tiered annotation (Je final_v2), 4 ep whole-step, Lyra 117139 (V2 1.7B, RESULTS §2i decision pair)

Source: RCP search-lm/ckpts/wiki6m/search_1p7B_tier_v2_fast (weights + tokenizer + trainerstate only; optimizer states and intermediate `checkpoint-*` dirs not uploaded). Pushed 2026-09-14. Numbers and protocol: Search-LLM `RESULTS.md` / `EXPERIMENTPLAN.md`.

field	value
epoch	4.0
global_step	30364
final_lr	2.2293622059521566e-11
final_loss	1.4988
train_loss	1.2785886130209199
safetensors_bytes	6845563296