model card

README.md

@@ -0,0 +1,119 @@
+---
+license: cc-by-nc-sa-4.0
+language:
+- en
+tags:
+  - world-model
+  - video-generation
+  - rocket-league
+  - interactive
+library_name: mira
+pipeline_tag: other
+---
+
+# Model Card for MIRA Mini PSD
+
+The few-step variant of MIRA Mini: the same 1B action-conditioned world model of
+Rocket League, distilled so that two sampling steps replace eight. Roughly three
+times the frame rate of the base model on the same hardware, at matched visual
+quality on our evaluations.
+
+> Built on [MIRA](https://mira-wm.com/), released July 6, 2026 by
+> [General Intuition](https://www.generalintuition.com/) and [Kyutai](https://kyutai.org/)
+> with Epic Games: [code](https://github.com/mira-wm/mira),
+> [dataset](https://huggingface.co/datasets/kyutai/rocket-science), and a published training
+> recipe. MIRA Mini is Alakazam's independent reproduction and optimization of that work.
+
+## Model Details
+
+### Model Description
+
+Identical architecture to [alakazamworld/mira-mini](https://huggingface.co/alakazamworld/mira-mini)
+(a 1B diffusion transformer in the latent space of a representation-autoencoder codec),
+plus the paper's progressive self-distillation (PSD), applied post hoc. Sampling a latent
+frame integrates the flow-matching field; PSD distills the model so one large integration
+step replaces two smaller ones, which lets two steps reach the quality that previously
+took eight.
+
+The paper distills its pretrained model and leaves unspecified how the step-size (Δ)
+conditioning pathway is introduced into a checkpoint trained without one. This model uses
+our construction: the pathway is injected with a zero-initialized output projection, so at
+step 0 the model is identical to the base checkpoint, and the distillation signal fades in
+from there (10k steps, learning rate 3e-5, warmup 100 — the from-scratch learning rate
+diverges on a converged checkpoint).
+
+- **Developed by:** Alakazam
+- **Model type:** Action-conditioned world model (interactive video generation)
+- **License:** CC BY-NC-SA 4.0, inherited from the training dataset
+- **Reproduction of:** MIRA (General Intuition and Kyutai, with Epic Games)
+
+**This model is for demonstration and research only.** The training dataset
+([kyutai/rocket-science](https://huggingface.co/datasets/kyutai/rocket-science)) is
+CC BY-NC-SA 4.0, with Rocket League content used by Epic Games' permission. These weights
+inherit that license: non-commercial, share-alike, with attribution.
+
+### Model Sources
+
+- **Player (one command):** [Alakazam-studios/alakazam-mira](https://github.com/Alakazam-studios/alakazam-mira)
+- **Technical report:** [alakazam.gg/mira](https://alakazam.gg/mira)
+- **Base 1B model:** [alakazamworld/mira-mini](https://huggingface.co/alakazamworld/mira-mini)
+- **Upstream release:** [mira-wm/mira](https://github.com/mira-wm/mira)
+
+## How to Get Started
+
+```bash
+pip install alakazam-mira
+mira play --model mira-mini-psd
+```
+
+The bundle is `world_model_config.yaml`, `checkpoint-10000/checkpoint.pth`, `codec/`,
+`context/default.npz`. Two steps is the intended setting; 4 and 8 also work (quality is
+flat across them after distillation, so extra steps mostly cost frame rate).
+
+## Training Details
+
+Post-hoc PSD fine-tune of the mira-mini 1B checkpoint: 10k steps on 8 preemptible H100s,
+PSD loss applied stochastically on 10% of updates (the paper's mixing rate), ground-truth
+flow term otherwise. Validation loss improved from 0.3952 (base, at handoff) to 0.3741,
+and the 2-step teacher-forced LPIPS ends below the base model's 2-step LPIPS with the
+margin still growing at 10k (−0.0021 on our held-out shard).
+
+Training data (through the base model): [kyutai/rocket-science](https://huggingface.co/datasets/kyutai/rocket-science),
+as released, no additions.
+
+## Performance
+
+| Hardware | 8 steps (base) | 2 steps (this model) |
+|---|---|---|
+| L40S | 14.5 fps | 19.4 fps measured (live serving probe) |
+| B200 | 25.7 fps | ~40 to 50 fps (est) |
+| L4 | 5.4 fps | ~20 to 25 fps (est) |
+
+Measured numbers come from serving probes on the stated hardware; estimates are marked.
+The serving runtime is the FlashDreams CUDA-graph port (bit-exact against the reference
+implementation at matched settings).
+
+## Bias, Risks, and Limitations
+
+- Same limitations as the base model: bot-collected Rocket League only, no transfer,
+  plausible continuations rather than exact physics.
+- Few-step sampling drifts slightly more per step than 8-step sampling on long rollouts;
+  quantitatively small on our evals, occasionally visible as texture softening.
+- Non-commercial license, inherited from the dataset.
+
+## Citation
+
+Cite the MIRA paper and link this repository.
+
+```bibtex
+@article{hu2026mira,
+  title  = {Multiplayer Interactive World Models with Representation Autoencoders},
+  author = {Hu, Anthony and others},
+  year   = {2026},
+  note   = {arXiv:2607.05352}
+}
+```
+
+## Model Card Authors
+
+Alakazam (alakazam.gg)