CoolFace
Modelpublic

zhiqilicg/flowmap_grpo_sCM

sourceHugging Faceotherupdated 13h agoView on Hugging Face
0likes
Model Card

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)

FolderFamilyRewardW&B runCheckpointsSection
checkpoints/meanflow/geneval/flowmap-grpo-r793cbvmMeanFlowGenEvalr793cbvmcheckpoint-1000E2
checkpoints/scm/geneval/flowmap-grpo-pt1j8w8gsCMGenEval (beta=0)pt1j8w8gcheckpoint-1000E5
checkpoints/meanflow/ocr/flowmap-grpo-zswhtiz7MeanFlowOCRzswhtiz7checkpoint-800E7
checkpoints/scm/ocr/flowmap-grpo-7j9ph5vfsCMOCR7j9ph5vfcheckpoint-700E6
checkpoints/meanflow/pickscore/flowmap-grpo-xw65jntzMeanFlowPickScore (1500 steps)xw65jntzcheckpoint-{300,1500}E9 (ckpt-1500), E0 (ckpt-300)
checkpoints/scm/pickscore/flowmap-grpo-h6j29iddsCMPickScore (1500 steps)h6j29iddcheckpoint-1500E8
checkpoints/meanflow/mix_reward_ocr70_pick30_ocrprompt/flowmap-grpo-rgawtwu8MeanFlowOCR 0.7 + PickScore 0.3, round 1 (OCR prompts only)rgawtwu8checkpoint-1000E3
checkpoints/scm/mix_reward_ocr70_pick30_ocrprompt/flowmap-grpo-7gce7didsCMOCR 0.7 + PickScore 0.3, round 1 (OCR prompts only)7gce7didcheckpoint-1000E4
checkpoints/meanflow/mix_reward_ocr70_pick30_mixprompt50/flowmap-grpoMeanFlowOCR 0.7 + PickScore 0.3, round 2 (OCR + PickScore prompts 50/50)nva0yjg0checkpoint-{950,1000}E12
checkpoints/scm/mix_reward_ocr70_pick30_mixprompt50/flowmap-grposCMOCR 0.7 + PickScore 0.3, round 2 (OCR + PickScore prompts 50/50)i4ukn52echeckpoint-{950,1000}E13

Other runs in this repo (longer Flow-Map GRPO runs and SFT / NFT baselines)

FolderCheckpoints
checkpoints/meanflow/ocr/flowmap-grpocheckpoint-1460..1500 (5 ckpts)
checkpoints/meanflow/pickscore/flowmap-grpocheckpoint-2460..2500 (5 ckpts)
checkpoints/scm/ocr/diffusion-nftcheckpoint-960..1000 (5 ckpts)
checkpoints/scm/ocr/flowmap-grpocheckpoint-1460..1500 (5 ckpts)
checkpoints/scm/ocr/online-sftcheckpoint-960..1000 (5 ckpts)
checkpoints/scm/pickscore/diffusion-nftcheckpoint-960..1000 (5 ckpts)
checkpoints/scm/pickscore/flowmap-grpocheckpoint-2460..2500 (5 ckpts)
checkpoints/scm/pickscore/online-sftcheckpoint-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.