Buckets:
| # Visual-OPSD Sanity Check. | |
| # Quick overfit on small data to verify pipeline works. | |
| # Expected: ~5 minutes. | |
| SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" | |
| ROOT_DIR="$(cd "${SCRIPT_DIR}/../.." && pwd)" | |
| cd "${ROOT_DIR}" | |
| export PYTHONPATH="${ROOT_DIR}:${PYTHONPATH:-}" | |
| source .venv/bin/activate | |
| MODEL_PATH=${1:-models/ThinkMorph-7B} | |
| echo "=== R001: SFT Sanity Check ===" | |
| torchrun \ | |
| --nproc_per_node=4 \ | |
| --master_port=29502 \ | |
| scripts/visual_opsd/train_visual_opsd_offline.py \ | |
| --mode sft \ | |
| --model_path ${MODEL_PATH} \ | |
| --dataset_config_file ./data/configs/text_reasoning.yaml \ | |
| --finetune_from_hf True \ | |
| --resume_model_only True \ | |
| --finetune_from_ema True \ | |
| --resume_from ${MODEL_PATH} \ | |
| --results_dir results/sanity_sft \ | |
| --checkpoint_dir results/sanity_sft/checkpoints \ | |
| --lr 1e-4 \ | |
| --num_workers 2 \ | |
| --max_num_tokens 4096 \ | |
| --visual_gen False \ | |
| --total_steps 50 \ | |
| --log_every 5 \ | |
| --save_every 100 \ | |
| --num_shard 4 \ | |
| --expected_num_tokens 4096 \ | |
| --warmup_steps 5 \ | |
| --wandb_offline True \ | |
| --wandb_project Visual-OPSD-sanity \ | |
| --wandb_name sanity-sft \ | |
| --kd_weight 0.0 | |
| echo "" | |
| echo "=== Sanity check complete. Check loss convergence in results/sanity_sft/ ===" | |
Xet Storage Details
- Size:
- 1.24 kB
- Xet hash:
- 753122413647e496d947164fa40fa48bafdf0ca72de60db81ed7eeb1ebcb67e6
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.