BELT-Qwen3-1.7B

Code and reproduction guide

Original trained artifacts for Same Beliefs, Different Policies: Learning Reusable Representations for Early Stopping in LLM Reasoning. Files retain their original bytes. CHECKPOINTS.json records the seed, training price where applicable, original source path, byte count and SHA256. Each run includes the original summary.json and available execution metadata. SHA256SUMS covers all supplied files except itself.

These are custom reader/head checkpoints used alongside the pinned base language model, loaded through the BELT repository's scorer. They are not standalone language models. No backbone weights, question manifests, full trajectory banks or fitted BELT stopping policies are bundled. See the code repository's recipes and asset list.

Source attribution is in the repository's NOTICE. This artifact publication assigns no new project license; backbone and upstream terms continue to apply.

Backbone

Qwen/Qwen3-1.7B at revision 70d244cc86ccca08cf5af4e1e306ecf908b1ad5e.

Contents and use

Three seeds (0, 1, 2), each with s1_best.pt, contain the four-target reader and readouts (q,v,r,stability = Now, End, Later, Same). Use a run directory as --source-run with iclr_score_pool; fit or obtain the separate stopping-policy parameters before policy evaluation.

This is the random-initialized family used by the main Qwen tables. The trainer loaded a layer-16 ReProbe template and used --random-init-head. The exact required template is included under reader-template/ (configuration and initialization weights, not an additional BELT result). Pass that directory as --head-checkpoint. Use the root Qwen runtime.

Each run also includes its original val_scored.jsonl. The scorer requires this file and copies it into the scored output for calibration; row counts, byte counts and hashes are recorded under validation_scores in CHECKPOINTS.json. These records are the saved validation scores, not newly generated data.

Checkpoint inventory

Directory Seed Training price Role SHA256
seed-0 0 β€” BELT primary ef0efb23afa73a2882706c1ef3277da13e0803bcf49255388ffaf5c8e4da4474
seed-1 1 β€” BELT primary 2013a17f7ebc0e3313df4e8815382009627064d29d1bbe97df03a4b173d59f7d
seed-2 2 β€” BELT primary d595fb20408d3b05495235b38a270bbc9c6a9e37c7b5ce3f73962893095b4835
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. πŸ™‹ Ask for provider support

Model tree for JingweiNi/BELT-Qwen3-1.7B

Finetuned
Qwen/Qwen3-1.7B
Finetuned
(1,424)
this model