CoolFace
Modelpublic

Falln87/clerk-memory

sourceHugging Faceapache-2.0updated 13d agoView on Hugging Face
0likes
25 commits on main
76a8e7d13d ago

CLERK paper draft v0.1: method, benchmark spec, protocol; results tables pending observed runs

Falln87
7aab36513d ago

CLERK project README: method summary, novelty statement, repro instructions

Falln87
1c428b713d ago

CLERK end-to-end reproducible pipeline (CPU smoke tests -> data -> LoRA SFT -> both evals)

Falln87
920a0c613d ago

eval_locomo: fix choice_logprobs continuation scoring

Falln87
705d9d413d ago

eval_locomo: clean rewrite (working choice log-likelihood scorer, consistent templates, single model loader)

Falln87
ca30e3513d ago

CLERK LoCoMo-MC10 zero-shot eval: incremental ledger build over LoCoMo sessions, MC log-likelihood scoring, full-context + prompted-ledger + summary-RAG baselines

Falln87
1d30dff13d ago

generator: fix gold_ops_for_events double-application bug; single-state implementation

Falln87
76e7c0913d ago

tests_smoke: clean rewrite of invariant checks (gold ops executable, budget, round-trip, QA consistency)

Falln87
c88c20f13d ago

eval_synthetic: initialize stale-error counters in Runner.__init__

Falln87
79f4cda13d ago

CLERK pinned dependencies

Falln87
415817c13d ago

CLERK smoke tests: gold ops executable, ledger round-trip, budget enforcement, QA supervision consistency

Falln87
f8cde5f13d ago

eval_synthetic: full clean rewrite (consistent generate contract, main, serialization helper)

Falln87
193e1ae13d ago

CLERK synthetic eval: clerk rollout + prompted-ledger/full-context/sliding-window/RAG baselines, per-type accuracy, memory-state metrics, token costs

Falln87
638039013d ago

train_sft: honor --base-model and --max-length, drop unused import

Falln87
4d8daa413d ago

CLERK LoRA SFT trainer (TRL SFTTrainer), verified against trl examples/sft_qlora; pushes adapter to the Hub

Falln87
c49870613d ago

CLERK SFT dataset builder: timelines -> messages JSONL (CONSOLIDATE + ANSWER mixture), seeded shuffle

Falln87
865342a13d ago

CLERK shared prompt definitions (train/eval must never drift)

Falln87
9dae93113d ago

generator: temporal answer uses slot-held value; gold ADD evicts first when ledger full; budget default 12 for eviction pressure

Falln87
61cccdd13d ago

generator: remove module-global session index; question targets restricted to ledger-supported predicates

Falln87
a042d2813d ago

common.py: fix facts_of valid_only semantics

Falln87
d03cb0b13d ago

CLERK generator: correct single-pass timeline builder, ledger-aware question targets, eviction-safe gold ops

Falln87
3cd7b4e13d ago

CLERK generator: clean full rewrite of render_session and persona timeline builder

Falln87
fdcd0a413d ago

CLERK synthetic evolving-session generator: personas, updates/supersessions/tombstones, gold ledger transitions, seeded

Falln87
4f70a2613d ago

CLERK core: slot schema, edit-op schema, deterministic reducer, serializers

Falln87
36958cf13d ago

initial commit

Falln87