Flow-Map GRPO checkpoints (FLUX.1-lite MeanFlow / sCM students)
LoRA adapters (r=64, alpha=128, 12 attention/FF projections) trained with Flow-Map GRPO on the T2I-Distill MeanFlow (flux1lite_meanflow.pt) and sCM (flux1lite_scm.pt) students of Freepik/flux.1-lite-8B, plus the online-SFT / Diffusion-NFT baselines. Code: ZhiqiLi-CG/flowmap_grpo (branch and commit in each run folder's README).
Layout: checkpoints/<family>/<task>/<method>[-<wandb run id>]/checkpoint-<step>/{lora, lora_ema, training_state}. lora/ is the raw learner, lora_ema/ the EMA adapter used for all reported numbers, training_state/ the full resume point. Run folders carry a README.md and the training-time full_eval_metrics.jsonl where available.
Evaluated Flow-Map GRPO runs (Genesis cluster; offline step sweeps in experiment_results.md, sections in the last column)
| Folder | Family | Reward | W&B run | Checkpoints | Section |
|---|---|---|---|---|---|
checkpoints/meanflow/geneval/flowmap-grpo-r793cbvm |
MeanFlow | GenEval | r793cbvm | checkpoint-1000 | E2 |
checkpoints/scm/geneval/flowmap-grpo-pt1j8w8g |
sCM | GenEval (beta=0) | pt1j8w8g | checkpoint-1000 | E5 |
checkpoints/meanflow/ocr/flowmap-grpo-zswhtiz7 |
MeanFlow | OCR | zswhtiz7 | checkpoint-800 | E7 |
checkpoints/scm/ocr/flowmap-grpo-7j9ph5vf |
sCM | OCR | 7j9ph5vf | checkpoint-700 | E6 |
checkpoints/meanflow/pickscore/flowmap-grpo-xw65jntz |
MeanFlow | PickScore (1500 steps) | xw65jntz | checkpoint-{300,1500} | E9 (ckpt-1500), E0 (ckpt-300) |
checkpoints/scm/pickscore/flowmap-grpo-h6j29idd |
sCM | PickScore (1500 steps) | h6j29idd | checkpoint-1500 | E8 |
checkpoints/meanflow/mix_reward_ocr70_pick30_ocrprompt/flowmap-grpo-rgawtwu8 |
MeanFlow | OCR 0.7 + PickScore 0.3, round 1 (OCR prompts only) | rgawtwu8 | checkpoint-1000 | E3 |
checkpoints/scm/mix_reward_ocr70_pick30_ocrprompt/flowmap-grpo-7gce7did |
sCM | OCR 0.7 + PickScore 0.3, round 1 (OCR prompts only) | 7gce7did | checkpoint-1000 | E4 |
checkpoints/meanflow/ocr_ablation_midpoint_anchor/flowmap-grpo-adglu9mr |
MeanFlow | OCR, kernel ablation (midpoint-anchor two-leg post_flow_renoise) | adglu9mr | checkpoint-1000 | E14 |
checkpoints/meanflow/mix_reward_ocr70_pick30_mixprompt50/flowmap-grpo |
MeanFlow | OCR 0.7 + PickScore 0.3, round 2 (OCR + PickScore prompts 50/50) | nva0yjg0 | checkpoint-{950,1000} | E12 |
checkpoints/scm/mix_reward_ocr70_pick30_mixprompt50/flowmap-grpo |
sCM | OCR 0.7 + PickScore 0.3, round 2 (OCR + PickScore prompts 50/50) | i4ukn52e | checkpoint-{950,1000} | E13 |
Other runs in this repo (longer Flow-Map GRPO runs and SFT / NFT baselines)
| Folder | Checkpoints |
|---|---|
checkpoints/meanflow/ocr/flowmap-grpo |
checkpoint-1460..1500 (5 ckpts) |
checkpoints/meanflow/pickscore/flowmap-grpo |
checkpoint-2460..2500 (5 ckpts) |
checkpoints/scm/ocr/diffusion-nft |
checkpoint-960..1000 (5 ckpts) |
checkpoints/scm/ocr/flowmap-grpo |
checkpoint-1460..1500 (5 ckpts) |
checkpoints/scm/ocr/online-sft |
checkpoint-960..1000 (5 ckpts) |
checkpoints/scm/pickscore/diffusion-nft |
checkpoint-960..1000 (5 ckpts) |
checkpoints/scm/pickscore/flowmap-grpo |
checkpoint-2460..2500 (5 ckpts) |
checkpoints/scm/pickscore/online-sft |
checkpoint-960..1000 (5 ckpts) |
Evaluation protocol for the GRPO runs: K = sampling steps - 1 (K=4 is the 5-step training sampler), EMA adapter, OCR test set (1018 prompts), PickScore test set (2048), GenEval (553 x 4), DrawBench (200 prompts x 5) with PickScore / Aesthetic / DeQA / ImageReward / UnifiedReward; details and all numbers in the run READMEs and experiment_results.md.