メニュー
Trains mean-velocity models on rectified couplings from pretrained flow models to dramatically smooth loss landscape, enabling faster convergence and superior one-step generation quality without additional training data.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/ADu2021/skillXiv --skill meanflow-one-step-generationコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Convert arXiv papers into ready-to-use agent skills using category-aware extraction. First classifies the paper into one or more of 11 research categories, then applies a specialized extraction pipeline for each category — because different types of papers produce different types of usable knowledge. A single paper can yield multiple skills if it spans categories. Use this skill whenever the user wants to turn a paper into a skill, extract practical techniques from research, build a skill library from papers, convert arXiv papers into reusable agent instructions, or batch-process multiple papers into skills. Also trigger when someone asks about extracting actionable knowledge from papers, making research practical for LLM agents, or systematically converting academic contributions into structured agent capabilities.
Establish regret bounds for behavior cloning with discretized actions combining statistical error and quantization error terms. Prove smoothness requirements for safe quantizer design, show that learning-based quantizers fail these requirements, and propose model-based augmentation to reduce error dependence from H² to H.
Dynamically allocate LoRA ranks per-layer during fine-tuning instead of using fixed uniform ranks. Learn optimal rank for each layer and subject via variational framework with discretized exponential distribution, reducing memory footprint while maintaining fidelity and text-alignment.
SOC 職業分類に基づく
SKILL.md を表示中
| name | meanflow-one-step-generation |
| title | Flow Straighter and Faster: Efficient One-Step Generation via MeanFlow with Rectified Trajectories |
| version | 0.0.2 |
| engine | skillxiv-v0.0.2-claude-opus-4.6 |
| license | MIT |
| url | https://arxiv.org/abs/2511.23342 |
| keywords | ["flow-matching","one-step-generation","trajectory-learning","diffusion-distillation","generative-models"] |
| description | Trains mean-velocity models on rectified couplings from pretrained flow models to dramatically smooth loss landscape, enabling faster convergence and superior one-step generation quality without additional training data. |
Flow Straighter and Faster (MeanFlow) addresses one-step generation by identifying that learning mean-velocity fields on highly curved generative trajectories creates noisy optimization. The key innovation trains mean-velocity models on rectified couplings (straighter trajectories from pretrained flow models) combined with distance-based truncation heuristics, dramatically smoothing the loss landscape.
Trajectory Rectification: Instead of learning velocity from independent random couplings (which have arbitrary curvature), use a pretrained flow model to generate "straighter" trajectories. The pretrained model has already learned optimal transports, and sampling from its trajectories provides smoother data for the student model.
Distance-Based Truncation: Remove the most curved trajectory pairs using a distance metric. If two adjacent samples in the coupling show large ||x_end - x_start|| relative to ||z_end - z_start||, they're likely noisy and removed from training.
Mean-Velocity Learning: Train a lightweight model to predict average velocity along trajectories:
v_mean = E[dx/dt] along trajectory
This is simpler than predicting full trajectories and benefits from rectified data.
Pretrained flow generation: Load a trained flow matching model (e.g., stable diffusion flow). Sample trajectories for one-step student training.
Trajectory selection: For candidates in the batch, compute:
curvature = ||x_t - x_0|| / (t + ε) # Distance per unit time
if curvature > threshold:
skip_sample(trajectory)
Mean-velocity model: Train a small transformer:
velocity = mlp(concat(x_0, x_T, t)) # Predict average velocity
loss = mse(velocity, (x_T - x_0) / (t_max - t_min))
One-step inference: At test time, sample z ~ N(0,I) and apply: x = z + velocity(z, t=0), with optional refinement.