sync training files

sync training files

fix: OOM — NUM_GENERATIONS 32→16, max_completion_length 300→200, expandable_segments

fix: add 3 missing hard tasks to _TASK_DIFFICULTY (322 prompts)

Fix: kl_coeff -> beta in colab

Fix: kl_coeff -> beta (correct TRL GRPOConfig param name)

Auto-detect username from token for adapter + run folder upload

Fix colab: format reward +-0.15, temp=0.7, kl_coeff=0.1

Fix: start_training yield count (9), plt.close memory leak

Fix restart loop: variant=secondary, theme back in Blocks()

Stop button: write flag file, wait 120s for clean save, fallback terminate

Graceful stop: save weights on /app/stop_requested flag

Add Stop button + save plots on stop; fix _refresh unpack bug; fix theme deprecation

Fix hard_currency_conversion task ID in TRAIN_TASKS and EVAL_TASKS

Baseline: Qwen2.5-7B-Instruct untrained 2026-04-26_0443

Add oversight eval + bump seeds medium×8 hard/long×20

Bump seeds: medium×8, hard/long×20 (322 prompts total)

Full metrics in live JSON: format/diff/ep_len history

Add format/difficulty/ep-length live panels + full metrics in JSON

Add loss tracking callback + reward/loss PNG savers

Add live loss curve panel + wire into dashboard

UI: generations default 16→32, max 16→64

Run 3: temp=0.7, kl=0.1, format±0.15, no curriculum, 20 tasks, G=32

Fix bf16 crash + 17 tasks / 160 prompts dataset

Update UI defaults: epochs 3→6 (max 10), generations 8→16 (max 32)

Baseline: Qwen2.5-7B-Instruct untrained 2026-04-25_1626

Update root eval_baseline.py to 17 tasks + long-horizon + health retry

Restore original requirements: gradio + torch ML stack

Fix requirements: restore gradio + UI deps for training space

Upload requirements.txt with huggingface_hub

Upload requirements.txt with huggingface_hub

Upload requirements.txt with huggingface_hub

Upload training/train.py with huggingface_hub

Upload training/eval_baseline.py with huggingface_hub

Upload train.py with huggingface_hub

Upload train.py with huggingface_hub

Upload app.py with huggingface_hub

Upload eval_baseline.py with huggingface_hub

Upload train.py with huggingface_hub

Upload app.py with huggingface_hub

Upload train.py with huggingface_hub

Upload requirements.txt with huggingface_hub

Upload app.py with huggingface_hub

Upload train.py with huggingface_hub

Upload requirements.txt with huggingface_hub

Upload Dockerfile with huggingface_hub

Upload train.py with huggingface_hub

Upload requirements.txt with huggingface_hub

Upload requirements.txt with huggingface_hub

Upload app.py with huggingface_hub