-
Upload train/preflight.py with huggingface_hub
-
audit: vectorised rolling-hash grams + sorted-array lookup; the Python-set version could not finish 1.2B tokens
-
train: --grad-ckpt, and record attn/grad_ckpt in every result line
-
train: --benchmark N runs the same step with no Trainer, to attribute the P2 throughput gap
-
train: explicit attention kernel (default eager, per p0c on Turing); report requested vs effective
-
train: preflight -- static-review fixes: absolute-step cursor, hub_listing pagination, cursor required on resume, repo_type everywhere, rank-0 repo create
-
train: train_ounce100m -- static-review fixes: absolute-step cursor, hub_listing pagination, cursor required on resume, repo_type everywhere, rank-0 repo create
-
train: hubckpt -- static-review fixes: absolute-step cursor, hub_listing pagination, cursor required on resume, repo_type everywhere, rank-0 repo create
-
train: shard_dataset -- static-review fixes: absolute-step cursor, hub_listing pagination, cursor required on resume, repo_type everywhere, rank-0 repo create
-
preflight: P2 fast-fails after the first dead cell (each repeat bills a GPU session)
-
preflight: P2 parses RUN_JSON per cell and requires the frozen 22L/2048 cell to report
-
train: frozen-width keeps GQA 3:1; report peak_gpu_gb and per-run tok/s
-
shard_dataset: drop the removed drop_last param's leftover body check (NameError at first call)
-
preflight: assert the trapezoid SHAPE (off-by-one at the last step is expected, not a bug)
-
train: train_ounce100m -- readback counts as verification; GQA kv must divide heads; pargs reports before optional probe
-
train: preflight -- readback counts as verification; GQA kv must divide heads; pargs reports before optional probe
-
train: hubckpt -- readback counts as verification; GQA kv must divide heads; pargs reports before optional probe
-
train: train_ounce100m -- download_checkpoint prefix/landing-dir fix (found by P1 v2)
-
train: preflight -- download_checkpoint prefix/landing-dir fix (found by P1 v2)
-
train: hubckpt -- download_checkpoint prefix/landing-dir fix (found by P1 v2)
-
train: fix SwiGLU width in build_config variants (was 2x)
-
preflight: stage pargs -- construct TrainingArguments/model/schedule on the Kaggle image, cross-check 22L=106,194,240 and 20L=99,114,048
-
train: train_ounce100m -- hubckpt.ensure_repo writes LFS attributes before the first checkpoint push (found by P1)
-
train: hubckpt -- hubckpt.ensure_repo writes LFS attributes before the first checkpoint push (found by P1)
-
train: preflight -- hubckpt.ensure_repo writes LFS attributes before the first checkpoint push (found by P1)
-
train: state which tokenizer files shipped with the export
-
publish: ship tokenizer.json with the mix, sha-checked against the manifest
-
train: create the ckpt repo if absent before the first push
-
train: Gate 3 preflight driver (reader, hub cycle, T1 throughput, cold resume)
-
train: add train_ounce100m (Phase 3/4 glue)
-
train: add hubckpt (Phase 3/4 glue)
-
train: add shard_dataset (Phase 3/4 glue)
-
audit: composite AUDIT_PASSED verdict; missing shards are unmeasured, not clean
-
rehearsal: make the script self-contained re sys.path/cwd (E-021)
-
rehearsal: grab_json returns a reason, manifest fallback, single-line VERDICT
-
build_mix: no stray print inside BUILD_JSON block (E-019); manifest gains n_shards
-
download_missing must fetch record.json too (.bin-only filter broke restore); rehearsal pre-cleans stale test repo
-
add publish_mix: LFS-first upload, generated card + attribution, Hub-side verification
-
stage listing raises instead of returning {}; empty-repo case handled explicitly; verify refuses manifest-less PASS; +val holdout in merge
-
hubsync.ensure_repo must install creds first (real cause of the 401); correct E-017 diagnosis
-
build_mix: hub-init failure is fatal (not silent), empty merge exits 4; add token-scope probe
-
hub sync for staged sources + cold-resume rehearsal script
-
smoke PASSED (4/4 sources per shard, 40/40 boundaries); add contamination audit; raw target 1.23B = 23% headroom
-
build_mix v2: stage->merge interleave + doc offset table; verify_mix asserts source-mix floor
-
build_mix: fix doc_hash byte/hex slice (smoke test caught it); add build/verify_mix.py
-
Phase 2: build_mix -- resumable, cross-source shards, seeded intra-shuffle, mandatory en filter
-
Phase 2: source inventory probe + d576/49152 param candidates
-
add p1_push_bench: Hub upload/download rate at checkpoint sizes
-
add credential helper + Hub write/bulk-upload roundtrip probe
-
add p1_secret_verify: mounted-credential end-to-end auth test
