14B ReadAll second-stage local10/global25 evaluated BF16 inference export.
Archived on 2026-09-28 at the owner's request. Only evaluated checkpoints are included; calibration-only and rejected candidates are explicitly labelled.
| Location | Evaluation | Notes |
|---|---|---|
. | HQA56/112/224 | Parent RL15, local10/global25; 40 new optimizer updates, 640 source questions. Original full training checkpoint previously pruned. |
Load the root with AutoTokenizer/AutoModelForCausalLM. Long-context evaluation requires the recurrent ReadTwice all-chunk loop, native context 32768, chunk size 5000, SKIM/UPDATE/FINAL budgets 512/1024/1024, temperature 0, top_p 1, seed 801. Parent model: Xirui1208/readall-readtwice-14b-rl-step15-20260925. Training reference: Xirui1208/readall-readtwice-14b-stage-a-step230-20260923. This is local10/global25, distinct from parent RL15 and local15/global30. No optimizer state is available for local10.
See archive_manifest.json for checkpoint provenance, evaluation_records/ for available reports, and sha256_manifest.json for file integrity.