QwenJev v0.4.3: 25-public-games eval in-repo (run_eval incremental report, fetch_games, CPU feasibility verified)

QwenJev v0.4.3: 25-public-games eval in-repo (run_eval incremental report, fetch_games, CPU feasibility verified)

QwenJev v0.4.3: 25-public-games eval in-repo (run_eval incremental report, fetch_games, CPU feasibility verified)

Upload ops/random_baseline_runfile.py with huggingface_hub

Upload ops/v5_daemon.py with huggingface_hub

Upload ops/teach_server.py with huggingface_hub

Upload ops/random_baseline_runfile.py with huggingface_hub

v0.4.10 fix2: REAL25_YAML env-aware (out_dir/dataset_dir) — hardcode trained v4b into v4 dir on old data; QWENJEV_DATASET env added

v0.4.10: heads-WIP upload hole (FUKJA61G lesson) + recover fixes (archive root, shutil.move nest, stamps prep/data only, max_steps 110)

v0.4.9: vision-aware pooling (RCA fix) + v4 namespace + eps0.45 collect config

v0.4.9: vision-aware pooling (RCA fix) + v4 namespace + eps0.45 collect config

v0.4.8: safetensors-loss fix + HEAD verification + v0.4.7 OOM-gate sync

v0.4.8: safetensors-loss fix + HEAD verification + v0.4.7 OOM-gate sync

QwenJev v0.4.3: 25-public-games eval in-repo (run_eval incremental report, fetch_games, CPU feasibility verified)

v3: real-games training round tooling

v3: real-games training round tooling

QwenJev v0.4.3: 25-public-games eval in-repo (run_eval incremental report, fetch_games, CPU feasibility verified)

ops/run_eval: fix local_dir must be repo-level dir (sub-path doubling bug, caught in live test)

ops/run_eval: fix hf: path resolution (keep sub-path slashes — files land where hf_hub_download puts them)

ops/run_eval: clearer arc-agi setup hint in preflight error

ops/README: run_eval.py entry (v0.4.1)

ops: one-command eval entry (checkpoint resolve + adapter reassembly + auto backend + provenance report)

ops/: add executor + adapter sharder + note pusher (sync complete)

ops/kaggle_kernel: ready-to-push submission bundle

ops/: T4 pipeline toolbox (pipeline/scale/executor/sharder/smoke/NVRTC tar/README)