实现了Context 预计算和缓存功能，提升了采样效率。 psnr不下降

.claude/settings.local.json

@@ -0,0 +1,10 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(conda env list:*)",
+      "Bash(mamba env:*)",
+      "Bash(micromamba env list:*)",
+      "Bash(echo:*)"
+    ]
+  }
+}

.gitignore

@@ -55,7 +55,6 @@ coverage.xml
 *.pot
 
 # Django stuff:
-*.log
 local_settings.py
 db.sqlite3
 
@@ -121,10 +120,14 @@ localTest/
 fig/
 figure/
 *.mp4
-*.json
+
 Data/ControlVAE.yml
 Data/Misc
 Data/Pretrained
 Data/utils.py
 Experiment/checkpoint
 Experiment/log
+
+*.ckpt
+
+*.0

ckpts/LICENSE

@@ -0,0 +1,439 @@
+Attribution-NonCommercial-ShareAlike 4.0 International
+
+Copyright (c) 2016-2025 HangZhou YuShu TECHNOLOGY CO.,LTD. ("Unitree Robotics")
+
+=======================================================================
+
+Creative Commons Corporation ("Creative Commons") is not a law firm and
+does not provide legal services or legal advice. Distribution of
+Creative Commons public licenses does not create a lawyer-client or
+other relationship. Creative Commons makes its licenses and related
+information available on an "as-is" basis. Creative Commons gives no
+warranties regarding its licenses, any material licensed under their
+terms and conditions, or any related information. Creative Commons
+disclaims all liability for damages resulting from their use to the
+fullest extent possible.
+
+Using Creative Commons Public Licenses
+
+Creative Commons public licenses provide a standard set of terms and
+conditions that creators and other rights holders may use to share
+original works of authorship and other material subject to copyright
+and certain other rights specified in the public license below. The
+following considerations are for informational purposes only, are not
+exhaustive, and do not form part of our licenses.
+
+     Considerations for licensors: Our public licenses are
+     intended for use by those authorized to give the public
+     permission to use material in ways otherwise restricted by
+     copyright and certain other rights. Our licenses are
+     irrevocable. Licensors should read and understand the terms
+     and conditions of the license they choose before applying it.
+     Licensors should also secure all rights necessary before
+     applying our licenses so that the public can reuse the
+     material as expected. Licensors should clearly mark any
+     material not subject to the license. This includes other CC-
+     licensed material, or material used under an exception or
+     limitation to copyright. More considerations for licensors:
+    wiki.creativecommons.org/Considerations_for_licensors
+
+     Considerations for the public: By using one of our public
+     licenses, a licensor grants the public permission to use the
+     licensed material under specified terms and conditions. If
+     the licensor's permission is not necessary for any reason--for
+     example, because of any applicable exception or limitation to
+     copyright--then that use is not regulated by the license. Our
+     licenses grant only permissions under copyright and certain
+     other rights that a licensor has authority to grant. Use of
+     the licensed material may still be restricted for other
+     reasons, including because others have copyright or other
+     rights in the material. A licensor may make special requests,
+     such as asking that all changes be marked or described.
+     Although not required by our licenses, you are encouraged to
+     respect those requests where reasonable. More considerations
+     for the public:
+    wiki.creativecommons.org/Considerations_for_licensees
+
+=======================================================================
+
+Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International
+Public License
+
+By exercising the Licensed Rights (defined below), You accept and agree
+to be bound by the terms and conditions of this Creative Commons
+Attribution-NonCommercial-ShareAlike 4.0 International Public License
+("Public License"). To the extent this Public License may be
+interpreted as a contract, You are granted the Licensed Rights in
+consideration of Your acceptance of these terms and conditions, and the
+Licensor grants You such rights in consideration of benefits the
+Licensor receives from making the Licensed Material available under
+these terms and conditions.
+
+
+Section 1 -- Definitions.
+
+  a. Adapted Material means material subject to Copyright and Similar
+     Rights that is derived from or based upon the Licensed Material
+     and in which the Licensed Material is translated, altered,
+     arranged, transformed, or otherwise modified in a manner requiring
+     permission under the Copyright and Similar Rights held by the
+     Licensor. For purposes of this Public License, where the Licensed
+     Material is a musical work, performance, or sound recording,
+     Adapted Material is always produced where the Licensed Material is
+     synched in timed relation with a moving image.
+
+  b. Adapter's License means the license You apply to Your Copyright
+     and Similar Rights in Your contributions to Adapted Material in
+     accordance with the terms and conditions of this Public License.
+
+  c. BY-NC-SA Compatible License means a license listed at
+     creativecommons.org/compatiblelicenses, approved by Creative
+     Commons as essentially the equivalent of this Public License.
+
+  d. Copyright and Similar Rights means copyright and/or similar rights
+     closely related to copyright including, without limitation,
+     performance, broadcast, sound recording, and Sui Generis Database
+     Rights, without regard to how the rights are labeled or
+     categorized. For purposes of this Public License, the rights
+     specified in Section 2(b)(1)-(2) are not Copyright and Similar
+     Rights.
+
+  e. Effective Technological Measures means those measures that, in the
+     absence of proper authority, may not be circumvented under laws
+     fulfilling obligations under Article 11 of the WIPO Copyright
+     Treaty adopted on December 20, 1996, and/or similar international
+     agreements.
+
+  f. Exceptions and Limitations means fair use, fair dealing, and/or
+     any other exception or limitation to Copyright and Similar Rights
+     that applies to Your use of the Licensed Material.
+
+  g. License Elements means the license attributes listed in the name
+     of a Creative Commons Public License. The License Elements of this
+     Public License are Attribution, NonCommercial, and ShareAlike.
+
+  h. Licensed Material means the artistic or literary work, database,
+     or other material to which the Licensor applied this Public
+     License.
+
+  i. Licensed Rights means the rights granted to You subject to the
+     terms and conditions of this Public License, which are limited to
+     all Copyright and Similar Rights that apply to Your use of the
+     Licensed Material and that the Licensor has authority to license.
+
+  j. Licensor means the individual(s) or entity(ies) granting rights
+     under this Public License.
+
+  k. NonCommercial means not primarily intended for or directed towards
+     commercial advantage or monetary compensation. For purposes of
+     this Public License, the exchange of the Licensed Material for
+     other material subject to Copyright and Similar Rights by digital
+     file-sharing or similar means is NonCommercial provided there is
+     no payment of monetary compensation in connection with the
+     exchange.
+
+  l. Share means to provide material to the public by any means or
+     process that requires permission under the Licensed Rights, such
+     as reproduction, public display, public performance, distribution,
+     dissemination, communication, or importation, and to make material
+     available to the public including in ways that members of the
+     public may access the material from a place and at a time
+     individually chosen by them.
+
+  m. Sui Generis Database Rights means rights other than copyright
+     resulting from Directive 96/9/EC of the European Parliament and of
+     the Council of 11 March 1996 on the legal protection of databases,
+     as amended and/or succeeded, as well as other essentially
+     equivalent rights anywhere in the world.
+
+  n. You means the individual or entity exercising the Licensed Rights
+     under this Public License. Your has a corresponding meaning.
+
+
+Section 2 -- Scope.
+
+  a. License grant.
+
+       1. Subject to the terms and conditions of this Public License,
+          the Licensor hereby grants You a worldwide, royalty-free,
+          non-sublicensable, non-exclusive, irrevocable license to
+          exercise the Licensed Rights in the Licensed Material to:
+
+            a. reproduce and Share the Licensed Material, in whole or
+               in part, for NonCommercial purposes only; and
+
+            b. produce, reproduce, and Share Adapted Material for
+               NonCommercial purposes only.
+
+       2. Exceptions and Limitations. For the avoidance of doubt, where
+          Exceptions and Limitations apply to Your use, this Public
+          License does not apply, and You do not need to comply with
+          its terms and conditions.
+
+       3. Term. The term of this Public License is specified in Section
+          6(a).
+
+       4. Media and formats; technical modifications allowed. The
+          Licensor authorizes You to exercise the Licensed Rights in
+          all media and formats whether now known or hereafter created,
+          and to make technical modifications necessary to do so. The
+          Licensor waives and/or agrees not to assert any right or
+          authority to forbid You from making technical modifications
+          necessary to exercise the Licensed Rights, including
+          technical modifications necessary to circumvent Effective
+          Technological Measures. For purposes of this Public License,
+          simply making modifications authorized by this Section 2(a)
+          (4) never produces Adapted Material.
+
+       5. Downstream recipients.
+
+            a. Offer from the Licensor -- Licensed Material. Every
+               recipient of the Licensed Material automatically
+               receives an offer from the Licensor to exercise the
+               Licensed Rights under the terms and conditions of this
+               Public License.
+
+            b. Additional offer from the Licensor -- Adapted Material.
+               Every recipient of Adapted Material from You
+               automatically receives an offer from the Licensor to
+               exercise the Licensed Rights in the Adapted Material
+               under the conditions of the Adapter's License You apply.
+
+            c. No downstream restrictions. You may not offer or impose
+               any additional or different terms or conditions on, or
+               apply any Effective Technological Measures to, the
+               Licensed Material if doing so restricts exercise of the
+               Licensed Rights by any recipient of the Licensed
+               Material.
+
+       6. No endorsement. Nothing in this Public License constitutes or
+          may be construed as permission to assert or imply that You
+          are, or that Your use of the Licensed Material is, connected
+          with, or sponsored, endorsed, or granted official status by,
+          the Licensor or others designated to receive attribution as
+          provided in Section 3(a)(1)(A)(i).
+
+  b. Other rights.
+
+       1. Moral rights, such as the right of integrity, are not
+          licensed under this Public License, nor are publicity,
+          privacy, and/or other similar personality rights; however, to
+          the extent possible, the Licensor waives and/or agrees not to
+          assert any such rights held by the Licensor to the limited
+          extent necessary to allow You to exercise the Licensed
+          Rights, but not otherwise.
+
+       2. Patent and trademark rights are not licensed under this
+          Public License.
+
+       3. To the extent possible, the Licensor waives any right to
+          collect royalties from You for the exercise of the Licensed
+          Rights, whether directly or through a collecting society
+          under any voluntary or waivable statutory or compulsory
+          licensing scheme. In all other cases the Licensor expressly
+          reserves any right to collect such royalties, including when
+          the Licensed Material is used other than for NonCommercial
+          purposes.
+
+
+Section 3 -- License Conditions.
+
+Your exercise of the Licensed Rights is expressly made subject to the
+following conditions.
+
+  a. Attribution.
+
+       1. If You Share the Licensed Material (including in modified
+          form), You must:
+
+            a. retain the following if it is supplied by the Licensor
+               with the Licensed Material:
+
+                 i. identification of the creator(s) of the Licensed
+                    Material and any others designated to receive
+                    attribution, in any reasonable manner requested by
+                    the Licensor (including by pseudonym if
+                    designated);
+
+                ii. a copyright notice;
+
+               iii. a notice that refers to this Public License;
+
+                iv. a notice that refers to the disclaimer of
+                    warranties;
+
+                 v. a URI or hyperlink to the Licensed Material to the
+                    extent reasonably practicable;
+
+            b. indicate if You modified the Licensed Material and
+               retain an indication of any previous modifications; and
+
+            c. indicate the Licensed Material is licensed under this
+               Public License, and include the text of, or the URI or
+               hyperlink to, this Public License.
+
+       2. You may satisfy the conditions in Section 3(a)(1) in any
+          reasonable manner based on the medium, means, and context in
+          which You Share the Licensed Material. For example, it may be
+          reasonable to satisfy the conditions by providing a URI or
+          hyperlink to a resource that includes the required
+          information.
+       3. If requested by the Licensor, You must remove any of the
+          information required by Section 3(a)(1)(A) to the extent
+          reasonably practicable.
+
+  b. ShareAlike.
+
+     In addition to the conditions in Section 3(a), if You Share
+     Adapted Material You produce, the following conditions also apply.
+
+       1. The Adapter's License You apply must be a Creative Commons
+          license with the same License Elements, this version or
+          later, or a BY-NC-SA Compatible License.
+
+       2. You must include the text of, or the URI or hyperlink to, the
+          Adapter's License You apply. You may satisfy this condition
+          in any reasonable manner based on the medium, means, and
+          context in which You Share Adapted Material.
+
+       3. You may not offer or impose any additional or different terms
+          or conditions on, or apply any Effective Technological
+          Measures to, Adapted Material that restrict exercise of the
+          rights granted under the Adapter's License You apply.
+
+
+Section 4 -- Sui Generis Database Rights.
+
+Where the Licensed Rights include Sui Generis Database Rights that
+apply to Your use of the Licensed Material:
+
+  a. for the avoidance of doubt, Section 2(a)(1) grants You the right
+     to extract, reuse, reproduce, and Share all or a substantial
+     portion of the contents of the database for NonCommercial purposes
+     only;
+
+  b. if You include all or a substantial portion of the database
+     contents in a database in which You have Sui Generis Database
+     Rights, then the database in which You have Sui Generis Database
+     Rights (but not its individual contents) is Adapted Material,
+     including for purposes of Section 3(b); and
+
+  c. You must comply with the conditions in Section 3(a) if You Share
+     all or a substantial portion of the contents of the database.
+
+For the avoidance of doubt, this Section 4 supplements and does not
+replace Your obligations under this Public License where the Licensed
+Rights include other Copyright and Similar Rights.
+
+
+Section 5 -- Disclaimer of Warranties and Limitation of Liability.
+
+  a. UNLESS OTHERWISE SEPARATELY UNDERTAKEN BY THE LICENSOR, TO THE
+     EXTENT POSSIBLE, THE LICENSOR OFFERS THE LICENSED MATERIAL AS-IS
+     AND AS-AVAILABLE, AND MAKES NO REPRESENTATIONS OR WARRANTIES OF
+     ANY KIND CONCERNING THE LICENSED MATERIAL, WHETHER EXPRESS,
+     IMPLIED, STATUTORY, OR OTHER. THIS INCLUDES, WITHOUT LIMITATION,
+     WARRANTIES OF TITLE, MERCHANTABILITY, FITNESS FOR A PARTICULAR
+     PURPOSE, NON-INFRINGEMENT, ABSENCE OF LATENT OR OTHER DEFECTS,
+     ACCURACY, OR THE PRESENCE OR ABSENCE OF ERRORS, WHETHER OR NOT
+     KNOWN OR DISCOVERABLE. WHERE DISCLAIMERS OF WARRANTIES ARE NOT
+     ALLOWED IN FULL OR IN PART, THIS DISCLAIMER MAY NOT APPLY TO YOU.
+
+  b. TO THE EXTENT POSSIBLE, IN NO EVENT WILL THE LICENSOR BE LIABLE
+     TO YOU ON ANY LEGAL THEORY (INCLUDING, WITHOUT LIMITATION,
+     NEGLIGENCE) OR OTHERWISE FOR ANY DIRECT, SPECIAL, INDIRECT,
+     INCIDENTAL, CONSEQUENTIAL, PUNITIVE, EXEMPLARY, OR OTHER LOSSES,
+     COSTS, EXPENSES, OR DAMAGES ARISING OUT OF THIS PUBLIC LICENSE OR
+     USE OF THE LICENSED MATERIAL, EVEN IF THE LICENSOR HAS BEEN
+     ADVISED OF THE POSSIBILITY OF SUCH LOSSES, COSTS, EXPENSES, OR
+     DAMAGES. WHERE A LIMITATION OF LIABILITY IS NOT ALLOWED IN FULL OR
+     IN PART, THIS LIMITATION MAY NOT APPLY TO YOU.
+
+  c. The disclaimer of warranties and limitation of liability provided
+     above shall be interpreted in a manner that, to the extent
+     possible, most closely approximates an absolute disclaimer and
+     waiver of all liability.
+
+
+Section 6 -- Term and Termination.
+
+  a. This Public License applies for the term of the Copyright and
+     Similar Rights licensed here. However, if You fail to comply with
+     this Public License, then Your rights under this Public License
+     terminate automatically.
+
+  b. Where Your right to use the Licensed Material has terminated under
+     Section 6(a), it reinstates:
+
+       1. automatically as of the date the violation is cured, provided
+          it is cured within 30 days of Your discovery of the
+          violation; or
+
+       2. upon express reinstatement by the Licensor.
+
+     For the avoidance of doubt, this Section 6(b) does not affect any
+     right the Licensor may have to seek remedies for Your violations
+     of this Public License.
+
+  c. For the avoidance of doubt, the Licensor may also offer the
+     Licensed Material under separate terms or conditions or stop
+     distributing the Licensed Material at any time; however, doing so
+     will not terminate this Public License.
+
+  d. Sections 1, 5, 6, 7, and 8 survive termination of this Public
+     License.
+
+
+Section 7 -- Other Terms and Conditions.
+
+  a. The Licensor shall not be bound by any additional or different
+     terms or conditions communicated by You unless expressly agreed.
+
+  b. Any arrangements, understandings, or agreements regarding the
+     Licensed Material not stated herein are separate from and
+     independent of the terms and conditions of this Public License.
+
+
+Section 8 -- Interpretation.
+
+  a. For the avoidance of doubt, this Public License does not, and
+     shall not be interpreted to, reduce, limit, restrict, or impose
+     conditions on any use of the Licensed Material that could lawfully
+     be made without permission under this Public License.
+
+  b. To the extent possible, if any provision of this Public License is
+     deemed unenforceable, it shall be automatically reformed to the
+     minimum extent necessary to make it enforceable. If the provision
+     cannot be reformed, it shall be severed from this Public License
+     without affecting the enforceability of the remaining terms and
+     conditions.
+
+  c. No term or condition of this Public License will be waived and no
+     failure to comply consented to unless expressly agreed to by the
+     Licensor.
+
+  d. Nothing in this Public License constitutes or may be interpreted
+     as a limitation upon, or waiver of, any privileges and immunities
+     that apply to the Licensor or You, including from the legal
+     processes of any jurisdiction or authority.
+
+=======================================================================
+
+Creative Commons is not a party to its public
+licenses. Notwithstanding, Creative Commons may elect to apply one of
+its public licenses to material it publishes and in those instances
+will be considered the “Licensor.” The text of the Creative Commons
+public licenses is dedicated to the public domain under the CC0 Public
+Domain Dedication. Except for the limited purpose of indicating that
+material is shared under a Creative Commons public license or as
+otherwise permitted by the Creative Commons policies published at
+creativecommons.org/policies, Creative Commons does not authorize the
+use of the trademark "Creative Commons" or any other trademark or logo
+of Creative Commons without its prior written consent including,
+without limitation, in connection with any unauthorized modifications
+to any of its public licenses or any other arrangements,
+understandings, or agreements concerning use of licensed material. For
+the avoidance of doubt, this paragraph does not form part of the
+public licenses.
+
+Creative Commons may be contacted at creativecommons.org.

ckpts/README.md

@@ -0,0 +1,38 @@
+---
+tags:
+- robotics
+---
+
+# UnifoLM-WMA-0: A World-Model-Action (WMA) Framework under UnifoLM Family
+<p style="font-size: 1.2em;">
+    <a href="https://unigen-x.github.io/unifolm-world-model-action.github.io"><strong>Project Page</strong></a> | 
+    <a href="https://github.com/unitreerobotics/unifolm-world-model-action"><strong>Code</strong></a> |
+    <a href="https://huggingface.co/unitreerobotics/datasets"><strong>Dataset</strong></a> 
+  </p>
+<div align="center">
+  <div align="justify">
+    <b>UnifoLM-WMA-0</b> is Unitree‘s first open-source world-model–action architecture spanning multiple types of robotic embodiments, designed specifically for general-purpose robot learning. Its core component is a world-model capable of understanding the physical interactions between robots and the environments. This world-model provides two key functions: (a) <b>Simulation Engine</b> – operates as an interactive simulator to generate synthetic data for robot learning; (b) <b>Policy Enhancement</b> – connects with an action head and, by predicting future interaction processes with the world-model, further optimizes decision-making performance.
+  </div>
+</div>
+
+## 🦾 Real Robot Deployment
+| <img src="assets/real_z1_stackbox.gif" style="border:none;box-shadow:none;margin:0;padding:0;" /> | <img src="assets/real_dual_stackbox.gif" style="border:none;box-shadow:none;margin:0;padding:0;" /> |
+|:---:|:---:|
+| <img src="assets/real_cleanup_pencils.gif" style="border:none;box-shadow:none;margin:0;padding:0;" /> | <img src="assets/real_g1_pack_camera.gif" style="border:none;box-shadow:none;margin:0;padding:0;" /> |
+
+**Note: the top-right window shows the world model’s prediction of future environmental changes.**
+
+## License
+The model is released under the CC BY-NC-SA 4.0 license as found in the [LICENSE](https://huggingface.co/unitreerobotics/UnifoLM-WMA-0/blob/main/LICENSE). You are responsible for ensuring that your use of Unitree AI Models complies with all applicable laws.
+
+## Model Architecture
+![Demo](assets/world_model_interaction.gif)
+
+## Citation
+```
+@misc{unifolm-wma-0,
+  author = {Unitree},
+  title  = {UnifoLM-WMA-0: A World-Model-Action (WMA) Framework under UnifoLM Family},
+  year   = {2025},
+}
+```

configs/inference/world_model_interaction.yaml

@@ -222,7 +222,7 @@ data:
     test:
       target: unifolm_wma.data.wma_data.WMAData
       params:
-        data_dir: '/path/to/unifolm-world-model-action/examples/world_model_interaction_prompts'
+        data_dir: '/home/qhy/unifolm-world-model-action/examples/world_model_interaction_prompts'
         video_length: ${model.params.wma_config.params.temporal_length}
         frame_stride: 2
         load_raw_resolution: True

psnr_score_for_challenge.py

@@ -0,0 +1,89 @@
+import os
+import glob
+import numpy as np
+import json
+from argparse import ArgumentParser, ArgumentDefaultsHelpFormatter
+from tqdm import tqdm
+from moviepy.video.io.VideoFileClip import VideoFileClip
+import PIL.Image
+
+
+def calculate_psnr(img1, img2):
+    mse = np.mean((img1.astype(np.float64) - img2.astype(np.float64)) ** 2)
+    if mse == 0:
+        return float('inf')
+    max_pixel = 255.0
+    psnr = 20 * np.log10(max_pixel / np.sqrt(mse))
+    return psnr
+
+
+def process_video_psnr(gt_path, pred_path):
+    try:
+        clip_gt = VideoFileClip(gt_path)
+        clip_pred = VideoFileClip(pred_path)
+
+        fps = min(clip_gt.fps, clip_pred.fps)
+        duration = min(clip_gt.duration, clip_pred.duration)
+
+        time_points = np.arange(0, duration, 1.0 / fps)
+
+        video_psnrs = []
+
+        for t in time_points:
+            frame_gt = clip_gt.get_frame(t)
+            frame_pred = clip_pred.get_frame(t)
+
+            img_gt = PIL.Image.fromarray(frame_gt).resize((256, 256), PIL.Image.Resampling.BILINEAR)
+            img_pred = PIL.Image.fromarray(frame_pred).resize((256, 256), PIL.Image.Resampling.BILINEAR)
+
+            psnr = calculate_psnr(np.array(img_gt), np.array(img_pred))
+            video_psnrs.append(psnr)
+
+        clip_gt.close()
+        clip_pred.close()
+
+        return np.mean(video_psnrs) if video_psnrs else 0.0
+
+    except Exception as e:
+        print(f"Error processing {os.path.basename(gt_path)}: {e}")
+        return None
+
+
+def main():
+    parser = ArgumentParser(formatter_class=ArgumentDefaultsHelpFormatter)
+    parser.add_argument('--gt_video', type=str, required=True, help='path to reference videos')
+    parser.add_argument('--pred_video', type=str, required=True, help='path to pred videos')
+    parser.add_argument('--output_file', type=str, default=None, help='path to output file')
+    args = parser.parse_args()
+
+    if not os.path.exists(args.gt_video):
+        print(f"Error: GT video not found at {args.gt_video}")
+        return
+    if not os.path.exists(args.pred_video):
+        print(f"Error: Pred video not found at {args.pred_video}")
+        return
+
+    print(f"Comparing:\nRef: {args.gt_video}\nPred: {args.pred_video}")
+
+    v_psnr = process_video_psnr(args.gt_video, args.pred_video)
+
+    if v_psnr is not None:
+        print("-" * 30)
+        print(f"Video PSNR: {v_psnr:.4f} dB")
+        print("-" * 30)
+
+        if args.output_file:
+            result = {
+                "gt_video": args.gt_video,
+                "pred_video": args.pred_video,
+                "psnr": v_psnr
+            }
+            with open(args.output_file, 'w') as f:
+                json.dump(result, f, indent=4)
+            print(f"Result saved to {args.output_file}")
+    else:
+        print("Failed to calculate PSNR.")
+
+
+if __name__ == '__main__':
+    main()

pyproject.toml

@@ -19,13 +19,13 @@ dependencies = [
     "pytorch-lightning==1.9.3",
     "pyyaml==6.0",
     "setuptools==65.6.3",
-    "torch==2.3.1",
+    #"torch==2.3.1",
-    "torchvision==0.18.1",
+    #"torchvision==0.18.1",
     "tqdm==4.66.5",
     "transformers==4.40.1",
     "moviepy==1.0.3",
     "av==12.3.0",
-    "xformers==0.0.27",
+    #"xformers==0.0.27",
     "gradio==4.39.0",
     "timm==0.9.10",
     "scikit-learn==1.5.1",

scripts/evaluation/base_model_inference.py

@@ -16,6 +16,9 @@ from collections import OrderedDict
 from unifolm_wma.models.samplers.ddim import DDIMSampler
 from unifolm_wma.utils.utils import instantiate_from_config
 
+torch.backends.cuda.matmul.allow_tf32 = True
+torch.backends.cudnn.allow_tf32 = True
+
 
 def get_filelist(data_dir: str, postfixes: list[str]) -> list[str]:
     """

scripts/evaluation/real_eval_server.py

@@ -19,6 +19,9 @@ from fastapi.responses import JSONResponse
 from typing import Any, Dict, Optional, Tuple, List
 from datetime import datetime
 
+torch.backends.cuda.matmul.allow_tf32 = True
+torch.backends.cudnn.allow_tf32 = True
+
 from unifolm_wma.utils.utils import instantiate_from_config
 from unifolm_wma.models.samplers.ddim import DDIMSampler
 

scripts/evaluation/world_model_interaction.py

@@ -18,6 +18,9 @@ from collections import OrderedDict
 from torch import nn
 from eval_utils import populate_queues, log_to_tensorboard
 from collections import deque
+
+torch.backends.cuda.matmul.allow_tf32 = True
+torch.backends.cudnn.allow_tf32 = True
 from torch import Tensor
 from torch.utils.tensorboard import SummaryWriter
 from PIL import Image
@@ -327,7 +330,8 @@ def image_guided_synthesis_sim_mode(
         timestep_spacing: str = 'uniform',
         guidance_rescale: float = 0.0,
         sim_mode: bool = True,
-        **kwargs) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        decode_video: bool = True,
+        **kwargs) -> tuple[torch.Tensor | None, torch.Tensor, torch.Tensor]:
     """
     Performs image-guided video generation in a simulation-style mode with optional multimodal guidance (image, state, action, text).
 
@@ -350,10 +354,13 @@ def image_guided_synthesis_sim_mode(
         timestep_spacing (str): Timestep sampling method in DDIM sampler. Typically "uniform" or "linspace".
         guidance_rescale (float): Guidance rescaling factor to mitigate overexposure from classifier-free guidance.
         sim_mode (bool): Whether to perform world-model interaction or decision-making using the world-model.
+        decode_video (bool): Whether to decode latent samples to pixel-space video.
+            Set to False to skip VAE decode for speed when only actions/states are needed.
         **kwargs: Additional arguments passed to the DDIM sampler.
 
     Returns:
-        batch_variants (torch.Tensor): Predicted pixel-space video frames [B, C, T, H, W].
+        batch_variants (torch.Tensor | None): Predicted pixel-space video frames [B, C, T, H, W],
+            or None when decode_video=False.
         actions (torch.Tensor): Predicted action sequences [B, T, D] from diffusion decoding.
         states (torch.Tensor): Predicted state sequences [B, T, D] from diffusion decoding.
     """
@@ -406,6 +413,7 @@ def image_guided_synthesis_sim_mode(
     kwargs.update({"unconditional_conditioning_img_nonetext": None})
     cond_mask = None
     cond_z0 = None
+    batch_variants = None
     if ddim_sampler is not None:
         samples, actions, states, intermedia = ddim_sampler.sample(
             S=ddim_steps,
@@ -424,6 +432,7 @@ def image_guided_synthesis_sim_mode(
             guidance_rescale=guidance_rescale,
             **kwargs)
 
+        if decode_video:
             # Reconstruct from latent to pixel space
             batch_images = model.decode_first_stage(samples)
             batch_variants = batch_images
@@ -587,7 +596,8 @@ def run_inference(args: argparse.Namespace, gpu_num: int, gpu_no: int) -> None:
                     fs=model_input_fs,
                     timestep_spacing=args.timestep_spacing,
                     guidance_rescale=args.guidance_rescale,
-                    sim_mode=False)
+                    sim_mode=False,
+                    decode_video=not args.fast_policy_no_decode)
 
                 # Update future actions in the observation queues
                 for idx in range(len(pred_actions[0])):
@@ -645,6 +655,7 @@ def run_inference(args: argparse.Namespace, gpu_num: int, gpu_no: int) -> None:
                                                       observation)
 
                 # Save the imagen videos for decision-making
+                if pred_videos_0 is not None:
                     sample_tag = f"{args.dataset}-vid{sample['videoid']}-dm-fs-{fs}/itr-{itr}"
                     log_to_tensorboard(writer,
                                        pred_videos_0,
@@ -658,6 +669,7 @@ def run_inference(args: argparse.Namespace, gpu_num: int, gpu_no: int) -> None:
                                    fps=args.save_fps)
 
                 # Save the imagen videos for decision-making
+                if pred_videos_0 is not None:
                     sample_video_file = f'{video_save_dir}/dm/{fs}/itr-{itr}.mp4'
                     save_results(pred_videos_0.cpu(),
                                  sample_video_file,
@@ -794,6 +806,11 @@ def get_parser():
                         action='store_true',
                         default=False,
                         help="not using the predicted states as comparison")
+    parser.add_argument(
+        "--fast_policy_no_decode",
+        action='store_true',
+        default=False,
+        help="Speed mode: policy pass only predicts actions, skip policy video decode/log/save.")
     parser.add_argument("--save_fps",
                         type=int,
                         default=8,

scripts/trainer.py

@@ -11,6 +11,9 @@ from unifolm_wma.utils.utils import instantiate_from_config
 from unifolm_wma.utils.train import get_trainer_callbacks, get_trainer_logger, get_trainer_strategy
 from unifolm_wma.utils.train import set_logger, init_workspace, load_checkpoints, get_num_parameters
 
+torch.backends.cuda.matmul.allow_tf32 = True
+torch.backends.cudnn.allow_tf32 = True
+
 
 def get_parser(**parser_kwargs):
     parser = argparse.ArgumentParser(**parser_kwargs)

src/unifolm_wma/models/diffusion_head/conditional_unet1d.py

@@ -501,6 +501,10 @@ class ConditionalUnet1D(nn.Module):
         self.last_frame_only = last_frame_only
         self.horizon = horizon
 
+        # Context precomputation cache
+        self._global_cond_cache_enabled = False
+        self._global_cond_cache = {}
+
     def forward(self,
                 sample: torch.Tensor,
                 timestep: Union[torch.Tensor, float, int],
@@ -530,6 +534,10 @@ class ConditionalUnet1D(nn.Module):
         B, T, D = sample.shape
         if self.use_linear_act_proj:
             sample = self.proj_in_action(sample.unsqueeze(-1))
+            _gc_key = (cond['image'].data_ptr(), cond['agent_pos'].data_ptr())
+            if self._global_cond_cache_enabled and _gc_key in self._global_cond_cache:
+                global_cond = self._global_cond_cache[_gc_key]
+            else:
                 global_cond = self.obs_encoder(cond)
                 global_cond = rearrange(global_cond,
                                         '(b t) d -> b 1 (t d)',
@@ -538,6 +546,8 @@ class ConditionalUnet1D(nn.Module):
                 global_cond = repeat(global_cond,
                                      'b c d -> b (repeat c) d',
                                      repeat=T)
+                if self._global_cond_cache_enabled:
+                    self._global_cond_cache[_gc_key] = global_cond
         else:
             sample = einops.rearrange(sample, 'b h t -> b t h')
             sample = self.proj_in_horizon(sample)

src/unifolm_wma/models/samplers/ddim.py

@@ -6,6 +6,8 @@ from unifolm_wma.utils.diffusion import make_ddim_sampling_parameters, make_ddim
 from unifolm_wma.utils.common import noise_like
 from unifolm_wma.utils.common import extract_into_tensor
 from tqdm import tqdm
+from unifolm_wma.modules.attention import enable_cross_attn_kv_cache, disable_cross_attn_kv_cache
+from unifolm_wma.modules.networks.wma_model import enable_ctx_cache, disable_ctx_cache
 
 
 class DDIMSampler(object):
@@ -67,11 +69,12 @@ class DDIMSampler(object):
             ddim_timesteps=self.ddim_timesteps,
             eta=ddim_eta,
             verbose=verbose)
-        self.register_buffer('ddim_sigmas', ddim_sigmas)
+        # Ensure tensors are on correct device for efficient indexing
-        self.register_buffer('ddim_alphas', ddim_alphas)
+        self.register_buffer('ddim_sigmas', to_torch(torch.as_tensor(ddim_sigmas)))
-        self.register_buffer('ddim_alphas_prev', ddim_alphas_prev)
+        self.register_buffer('ddim_alphas', to_torch(torch.as_tensor(ddim_alphas)))
+        self.register_buffer('ddim_alphas_prev', to_torch(torch.as_tensor(ddim_alphas_prev)))
         self.register_buffer('ddim_sqrt_one_minus_alphas',
-                             np.sqrt(1. - ddim_alphas))
+                             to_torch(torch.as_tensor(np.sqrt(1. - ddim_alphas))))
         sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
             (1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) *
             (1 - self.alphas_cumprod / self.alphas_cumprod_prev))
@@ -241,9 +244,13 @@ class DDIMSampler(object):
 
         dp_ddim_scheduler_action.set_timesteps(len(timesteps))
         dp_ddim_scheduler_state.set_timesteps(len(timesteps))
+        ts = torch.empty((b, ), device=device, dtype=torch.long)
+        enable_cross_attn_kv_cache(self.model)
+        enable_ctx_cache(self.model)
+        try:
             for i, step in enumerate(iterator):
                 index = total_steps - i - 1
-            ts = torch.full((b, ), step, device=device, dtype=torch.long)
+                ts.fill_(step)
 
                 # Use mask to blend noised original latent (img_orig) & new sampled latent (img)
                 if mask is not None:
@@ -298,6 +305,9 @@ class DDIMSampler(object):
                     intermediates['pred_x0'].append(pred_x0)
                     intermediates['x_inter_action'].append(action)
                     intermediates['x_inter_state'].append(state)
+        finally:
+            disable_cross_attn_kv_cache(self.model)
+            disable_ctx_cache(self.model)
 
         return img, action, state, intermediates
 
@@ -325,10 +335,6 @@ class DDIMSampler(object):
                       guidance_rescale=0.0,
                       **kwargs):
         b, *_, device = *x.shape, x.device
-        if x.dim() == 5:
-            is_video = True
-        else:
-            is_video = False
 
         if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
             model_output, model_output_action, model_output_state = self.model.apply_model(
@@ -377,17 +383,11 @@ class DDIMSampler(object):
         sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
         sigmas = self.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
 
-        if is_video:
+        # Use 0-d tensors directly (already on device); broadcasting handles shape
-            size = (b, 1, 1, 1, 1)
+        a_t = alphas[index]
-        else:
+        a_prev = alphas_prev[index]
-            size = (b, 1, 1, 1)
+        sigma_t = sigmas[index]
-
+        sqrt_one_minus_at = sqrt_one_minus_alphas[index]
-        a_t = torch.full(size, alphas[index], device=device)
-        a_prev = torch.full(size, alphas_prev[index], device=device)
-        sigma_t = torch.full(size, sigmas[index], device=device)
-        sqrt_one_minus_at = torch.full(size,
-                                       sqrt_one_minus_alphas[index],
-                                       device=device)
 
         if self.model.parameterization != "v":
             pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
@@ -395,12 +395,8 @@ class DDIMSampler(object):
             pred_x0 = self.model.predict_start_from_z_and_v(x, t, model_output)
 
         if self.model.use_dynamic_rescale:
-            scale_t = torch.full(size,
+            scale_t = self.ddim_scale_arr[index]
-                                 self.ddim_scale_arr[index],
+            prev_scale_t = self.ddim_scale_arr_prev[index]
-                                 device=device)
-            prev_scale_t = torch.full(size,
-                                      self.ddim_scale_arr_prev[index],
-                                      device=device)
             rescale = (prev_scale_t / scale_t)
             pred_x0 *= rescale
 

src/unifolm_wma/modules/attention.py

@@ -98,6 +98,9 @@ class CrossAttention(nn.Module):
         self.text_context_len = text_context_len
         self.agent_state_context_len = agent_state_context_len
         self.agent_action_context_len = agent_action_context_len
+        self._kv_cache = {}
+        self._kv_cache_enabled = False
+
         self.cross_attention_scale_learnable = cross_attention_scale_learnable
         if self.image_cross_attention:
             self.to_k_ip = nn.Linear(context_dim, inner_dim, bias=False)
@@ -236,17 +239,42 @@ class CrossAttention(nn.Module):
         k_ip, v_ip, out_ip = None, None, None
         k_as, v_as, out_as = None, None, None
         k_aa, v_aa, out_aa = None, None, None
+        attn_mask_aa = None
 
+        h = self.heads
         q = self.to_q(x)
         context = default(context, x)
 
-        if self.image_cross_attention and not spatial_self_attn:
+        b, _, _ = q.shape
+        q = q.unsqueeze(3).reshape(b, q.shape[1], h, self.dim_head).permute(0, 2, 1, 3).reshape(b * h, q.shape[1], self.dim_head).contiguous()
+
+        def _reshape_kv(t):
+            return t.unsqueeze(3).reshape(b, t.shape[1], h, self.dim_head).permute(0, 2, 1, 3).reshape(b * h, t.shape[1], self.dim_head).contiguous()
+
+        use_cache = self._kv_cache_enabled and not spatial_self_attn
+        cache_hit = use_cache and len(self._kv_cache) > 0
+
+        if cache_hit:
+            k = self._kv_cache['k']
+            v = self._kv_cache['v']
+            k_ip = self._kv_cache.get('k_ip')
+            v_ip = self._kv_cache.get('v_ip')
+            k_as = self._kv_cache.get('k_as')
+            v_as = self._kv_cache.get('v_as')
+            k_aa = self._kv_cache.get('k_aa')
+            v_aa = self._kv_cache.get('v_aa')
+            attn_mask_aa = self._kv_cache.get('attn_mask_aa')
+        elif self.image_cross_attention and not spatial_self_attn:
             if context.shape[1] == self.text_context_len + self.video_length:
                 context_ins, context_image = context[:, :self.text_context_len, :], context[:,self.text_context_len:, :]
                 k = self.to_k(context)
                 v = self.to_v(context)
                 k_ip = self.to_k_ip(context_image)
                 v_ip = self.to_v_ip(context_image)
+                k, v = map(_reshape_kv, (k, v))
+                k_ip, v_ip = map(_reshape_kv, (k_ip, v_ip))
+                if use_cache:
+                    self._kv_cache = {'k': k, 'v': v, 'k_ip': k_ip, 'v_ip': v_ip}
             elif context.shape[1] == self.agent_state_context_len + self.text_context_len + self.video_length:
                 context_agent_state = context[:, :self.agent_state_context_len, :]
                 context_ins = context[:, self.agent_state_context_len:self.agent_state_context_len+self.text_context_len, :]
@@ -257,6 +285,11 @@ class CrossAttention(nn.Module):
                 v_ip = self.to_v_ip(context_image)
                 k_as = self.to_k_as(context_agent_state)
                 v_as = self.to_v_as(context_agent_state)
+                k, v = map(_reshape_kv, (k, v))
+                k_ip, v_ip = map(_reshape_kv, (k_ip, v_ip))
+                k_as, v_as = map(_reshape_kv, (k_as, v_as))
+                if use_cache:
+                    self._kv_cache = {'k': k, 'v': v, 'k_ip': k_ip, 'v_ip': v_ip, 'k_as': k_as, 'v_as': v_as}
             else:
                 context_agent_state = context[:, :self.agent_state_context_len, :]
                 context_agent_action = context[:, self.agent_state_context_len:self.agent_state_context_len+self.agent_action_context_len, :]
@@ -272,98 +305,78 @@ class CrossAttention(nn.Module):
                 k_aa = self.to_k_aa(context_agent_action)
                 v_aa = self.to_v_aa(context_agent_action)
 
-                attn_mask_aa = self._get_attn_mask_aa(x.shape[0],
+                k, v = map(_reshape_kv, (k, v))
+                k_ip, v_ip = map(_reshape_kv, (k_ip, v_ip))
+                k_as, v_as = map(_reshape_kv, (k_as, v_as))
+                k_aa, v_aa = map(_reshape_kv, (k_aa, v_aa))
+
+                attn_mask_aa_raw = self._get_attn_mask_aa(x.shape[0],
                                                           q.shape[1],
                                                           k_aa.shape[1],
-                                                      block_size=16).to(k_aa.device)
+                                                          block_size=16,
+                                                          device=k_aa.device)
+                attn_mask_aa = attn_mask_aa_raw.unsqueeze(1).repeat(1, h, 1, 1).reshape(
+                    b * h, attn_mask_aa_raw.shape[1], attn_mask_aa_raw.shape[2]).to(q.dtype)
+
+                if use_cache:
+                    self._kv_cache = {
+                        'k': k, 'v': v, 'k_ip': k_ip, 'v_ip': v_ip,
+                        'k_as': k_as, 'v_as': v_as, 'k_aa': k_aa, 'v_aa': v_aa,
+                        'attn_mask_aa': attn_mask_aa,
+                    }
         else:
             if not spatial_self_attn:
                 assert 1 > 2, ">>> ERROR: you should never go into here ..."
                 context = context[:, :self.text_context_len, :]
             k = self.to_k(context)
             v = self.to_v(context)
-
+            k, v = map(_reshape_kv, (k, v))
-        b, _, _ = q.shape
+            if use_cache:
-        q = q.unsqueeze(3).reshape(b, q.shape[1], self.heads, self.dim_head).permute(0, 2, 1, 3).reshape(b * self.heads, q.shape[1], self.dim_head).contiguous()
+                self._kv_cache = {'k': k, 'v': v}
         if k is not None:
-            k, v = map(
-                lambda t: t.unsqueeze(3).reshape(b, t.shape[
-                    1], self.heads, self.dim_head).permute(0, 2, 1, 3).reshape(
-                        b * self.heads, t.shape[1], self.dim_head).contiguous(),
-                (k, v),
-            )
             out = xformers.ops.memory_efficient_attention(q,
                                                           k,
                                                           v,
                                                           attn_bias=None,
                                                           op=None)
             out = (out.unsqueeze(0).reshape(
-                b, self.heads, out.shape[1],
+                b, h, out.shape[1],
                 self.dim_head).permute(0, 2, 1,
                                        3).reshape(b, out.shape[1],
-                                                  self.heads * self.dim_head))
+                                                  h * self.dim_head))
 
         if k_ip is not None:
-            # For image cross-attention
-            k_ip, v_ip = map(
-                lambda t: t.unsqueeze(3).reshape(b, t.shape[
-                    1], self.heads, self.dim_head).permute(0, 2, 1, 3).reshape(
-                        b * self.heads, t.shape[1], self.dim_head).contiguous(
-                        ),
-                (k_ip, v_ip),
-            )
             out_ip = xformers.ops.memory_efficient_attention(q,
                                                              k_ip,
                                                              v_ip,
                                                              attn_bias=None,
                                                              op=None)
             out_ip = (out_ip.unsqueeze(0).reshape(
-                b, self.heads, out_ip.shape[1],
+                b, h, out_ip.shape[1],
                 self.dim_head).permute(0, 2, 1,
                                        3).reshape(b, out_ip.shape[1],
-                                                  self.heads * self.dim_head))
+                                                  h * self.dim_head))
 
         if k_as is not None:
-            # For agent state cross-attention
-            k_as, v_as = map(
-                lambda t: t.unsqueeze(3).reshape(b, t.shape[
-                    1], self.heads, self.dim_head).permute(0, 2, 1, 3).reshape(
-                        b * self.heads, t.shape[1], self.dim_head).contiguous(
-                        ),
-                (k_as, v_as),
-            )
             out_as = xformers.ops.memory_efficient_attention(q,
                                                              k_as,
                                                              v_as,
                                                              attn_bias=None,
                                                              op=None)
             out_as = (out_as.unsqueeze(0).reshape(
-                b, self.heads, out_as.shape[1],
+                b, h, out_as.shape[1],
                 self.dim_head).permute(0, 2, 1,
                                        3).reshape(b, out_as.shape[1],
-                                                  self.heads * self.dim_head))
+                                                  h * self.dim_head))
+
         if k_aa is not None:
-            # For agent action cross-attention
-            k_aa, v_aa = map(
-                lambda t: t.unsqueeze(3).reshape(b, t.shape[
-                    1], self.heads, self.dim_head).permute(0, 2, 1, 3).reshape(
-                        b * self.heads, t.shape[1], self.dim_head).contiguous(
-                        ),
-                (k_aa, v_aa),
-            )
-
-            attn_mask_aa = attn_mask_aa.unsqueeze(1).repeat(1,self.heads,1,1).reshape(
-                    b * self.heads, attn_mask_aa.shape[1], attn_mask_aa.shape[2])
-            attn_mask_aa = attn_mask_aa.to(q.dtype)
-
             out_aa = xformers.ops.memory_efficient_attention(
                 q, k_aa, v_aa, attn_bias=attn_mask_aa, op=None)
-
             out_aa = (out_aa.unsqueeze(0).reshape(
-                b, self.heads, out_aa.shape[1],
+                b, h, out_aa.shape[1],
                 self.dim_head).permute(0, 2, 1,
                                        3).reshape(b, out_aa.shape[1],
-                                                  self.heads * self.dim_head))
+                                                  h * self.dim_head))
         if exists(mask):
             raise NotImplementedError
 
@@ -386,17 +399,43 @@ class CrossAttention(nn.Module):
 
         return self.to_out(out)
 
-    def _get_attn_mask_aa(self, b, l1, l2, block_size=16):
+    def _get_attn_mask_aa(self, b, l1, l2, block_size=16, device=None):
+        cache_key = (b, l1, l2, block_size)
+        if hasattr(self, '_attn_mask_aa_cache_key') and self._attn_mask_aa_cache_key == cache_key:
+            cached = self._attn_mask_aa_cache
+            if device is not None and cached.device != torch.device(device):
+                cached = cached.to(device)
+                self._attn_mask_aa_cache = cached
+            return cached
+
+        target_device = device if device is not None else 'cpu'
         num_token = l2 // block_size
-        start_positions = ((torch.arange(b) % block_size) + 1) * num_token
+        start_positions = ((torch.arange(b, device=target_device) % block_size) + 1) * num_token
-        col_indices = torch.arange(l2)
+        col_indices = torch.arange(l2, device=target_device)
         mask_2d = col_indices.unsqueeze(0) >= start_positions.unsqueeze(1)
         mask = mask_2d.unsqueeze(1).expand(b, l1, l2)
-        attn_mask = torch.zeros_like(mask, dtype=torch.float)
+        attn_mask = torch.zeros(b, l1, l2, dtype=torch.float, device=target_device)
         attn_mask[mask] = float('-inf')
+
+        self._attn_mask_aa_cache_key = cache_key
+        self._attn_mask_aa_cache = attn_mask
         return attn_mask
 
 
+def enable_cross_attn_kv_cache(module):
+    for m in module.modules():
+        if isinstance(m, CrossAttention):
+            m._kv_cache_enabled = True
+            m._kv_cache = {}
+
+
+def disable_cross_attn_kv_cache(module):
+    for m in module.modules():
+        if isinstance(m, CrossAttention):
+            m._kv_cache_enabled = False
+            m._kv_cache = {}
+
+
 class BasicTransformerBlock(nn.Module):
 
     def __init__(self,

src/unifolm_wma/modules/networks/wma_model.py

@@ -685,6 +685,10 @@ class WMAModel(nn.Module):
         self.action_token_projector = instantiate_from_config(
             stem_process_config)
 
+        # Context precomputation cache
+        self._ctx_cache_enabled = False
+        self._ctx_cache = {}
+
     def forward(self,
                 x: Tensor,
                 x_action: Tensor,
@@ -720,6 +724,10 @@ class WMAModel(nn.Module):
                                    repeat_only=False).type(x.dtype)
         emb = self.time_embed(t_emb)
 
+        _ctx_key = context.data_ptr()
+        if self._ctx_cache_enabled and _ctx_key in self._ctx_cache:
+            context = self._ctx_cache[_ctx_key]
+        else:
             bt, l_context, _ = context.shape
             if self.base_model_gen_only:
                 assert l_context == 77 + self.n_obs_steps * 16, ">>> ERROR Context dim 1 ..."  ## NOTE HANDCODE
@@ -772,6 +780,8 @@ class WMAModel(nn.Module):
                         context_img
                     ],
                                         dim=1)
+            if self._ctx_cache_enabled:
+                self._ctx_cache[_ctx_key] = context
 
         emb = emb.repeat_interleave(repeats=t, dim=0)
 
@@ -846,3 +856,30 @@ class WMAModel(nn.Module):
             s_y = torch.zeros_like(x_state)
 
         return y, a_y, s_y
+
+
+def enable_ctx_cache(model):
+    """Enable context precomputation cache on WMAModel and its action/state UNets."""
+    for m in model.modules():
+        if isinstance(m, WMAModel):
+            m._ctx_cache_enabled = True
+            m._ctx_cache = {}
+    # conditional_unet1d cache
+    from unifolm_wma.models.diffusion_head.conditional_unet1d import ConditionalUnet1D
+    for m in model.modules():
+        if isinstance(m, ConditionalUnet1D):
+            m._global_cond_cache_enabled = True
+            m._global_cond_cache = {}
+
+
+def disable_ctx_cache(model):
+    """Disable and clear context precomputation cache."""
+    for m in model.modules():
+        if isinstance(m, WMAModel):
+            m._ctx_cache_enabled = False
+            m._ctx_cache = {}
+    from unifolm_wma.models.diffusion_head.conditional_unet1d import ConditionalUnet1D
+    for m in model.modules():
+        if isinstance(m, ConditionalUnet1D):
+            m._global_cond_cache_enabled = False
+            m._global_cond_cache = {}

unitree_g1_pack_camera/case1/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_g1_pack_camera/case1"
+dataset="unitree_g1_pack_camera"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_g1_pack_camera/case1/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 6 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 11 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_g1_pack_camera/case1/world_model_interaction_prompts/unitree_g1_pack_camera.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+0,x,x,unitree_g1_pack_camera,mount camera,x,x,x,G1_Dex1,30

unitree_g1_pack_camera/case2/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_g1_pack_camera/case2"
+dataset="unitree_g1_pack_camera"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_g1_pack_camera/case2/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 6 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 11 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_g1_pack_camera/case2/world_model_interaction_prompts/unitree_g1_pack_camera.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+50,x,x,unitree_g1_pack_camera,mount camera,x,x,x,G1_Dex1,30

unitree_g1_pack_camera/case3/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_g1_pack_camera/case3"
+dataset="unitree_g1_pack_camera"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_g1_pack_camera/case3/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 6 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 11 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_g1_pack_camera/case3/world_model_interaction_prompts/unitree_g1_pack_camera.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+100,x,x,unitree_g1_pack_camera,mount camera,x,x,x,G1_Dex1,30

unitree_g1_pack_camera/case4/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_g1_pack_camera/case4"
+dataset="unitree_g1_pack_camera"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_g1_pack_camera/case4/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 6 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 11 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_g1_pack_camera/case4/world_model_interaction_prompts/unitree_g1_pack_camera.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+200,x,x,unitree_g1_pack_camera,mount camera,x,x,x,G1_Dex1,30

unitree_z1_dual_arm_cleanup_pencils/case1/output.log

@@ -0,0 +1,121 @@
+2026-02-10 15:38:28.973314: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
+2026-02-10 15:38:29.023024: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+2026-02-10 15:38:29.023070: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+2026-02-10 15:38:29.024393: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+2026-02-10 15:38:29.031901: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
+To enable the following instructions: AVX2 AVX512F AVX512_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
+2026-02-10 15:38:29.955454: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
+Global seed set to 123
+INFO:mainlogger:LatentVisualDiffusion: Running in v-prediction mode
+INFO:unifolm_wma.models.diffusion_head.conditional_unet1d:number of parameters: 5.010531e+08
+INFO:unifolm_wma.models.diffusion_head.conditional_unet1d:number of parameters: 5.010531e+08
+AE working on z of shape (1, 4, 32, 32) = 4096 dimensions.
+INFO:root:Loaded ViT-H-14 model config.
+DEBUG:urllib3.connectionpool:Starting new HTTPS connection (1): hf-mirror.com:443
+DEBUG:urllib3.connectionpool:https://hf-mirror.com:443 "HEAD /laion/CLIP-ViT-H-14-laion2B-s32B-b79K/resolve/main/open_clip_pytorch_model.bin HTTP/1.1" 302 0
+INFO:root:Loading pretrained ViT-H-14 weights (laion2b_s32b_b79k).
+INFO:root:Loaded ViT-H-14 model config.
+DEBUG:urllib3.connectionpool:https://hf-mirror.com:443 "HEAD /laion/CLIP-ViT-H-14-laion2B-s32B-b79K/resolve/main/open_clip_pytorch_model.bin HTTP/1.1" 302 0
+INFO:root:Loading pretrained ViT-H-14 weights (laion2b_s32b_b79k).
+>>> model checkpoint loaded.
+>>> Load pre-trained model ...
+INFO:root:***** Configing Data *****
+>>> unitree_z1_stackbox: 1 data samples loaded.
+>>> unitree_z1_stackbox: data stats loaded.
+>>> unitree_z1_stackbox: normalizer initiated.
+>>> unitree_z1_dual_arm_stackbox: 1 data samples loaded.
+>>> unitree_z1_dual_arm_stackbox: data stats loaded.
+>>> unitree_z1_dual_arm_stackbox: normalizer initiated.
+>>> unitree_z1_dual_arm_stackbox_v2: 1 data samples loaded.
+>>> unitree_z1_dual_arm_stackbox_v2: data stats loaded.
+>>> unitree_z1_dual_arm_stackbox_v2: normalizer initiated.
+>>> unitree_z1_dual_arm_cleanup_pencils: 1 data samples loaded.
+>>> unitree_z1_dual_arm_cleanup_pencils: data stats loaded.
+>>> unitree_z1_dual_arm_cleanup_pencils: normalizer initiated.
+>>> unitree_g1_pack_camera: 1 data samples loaded.
+>>> unitree_g1_pack_camera: data stats loaded.
+>>> unitree_g1_pack_camera: normalizer initiated.
+>>> Dataset is successfully loaded ...
+>>> Generate 16 frames under each generation ...
+DEBUG:h5py._conv:Creating converter from 3 to 5
+DEBUG:PIL.PngImagePlugin:STREAM b'IHDR' 16 13
+DEBUG:PIL.PngImagePlugin:STREAM b'pHYs' 41 9
+DEBUG:PIL.PngImagePlugin:STREAM b'IDAT' 62 4096
+
+  0%|          | 0/8 [00:00<?, ?it/s]>>> Step 0: generating actions ...
+>>> Step 0: interacting with world model ...
+DEBUG:PIL.Image:Importing BlpImagePlugin
+DEBUG:PIL.Image:Importing BmpImagePlugin
+DEBUG:PIL.Image:Importing BufrStubImagePlugin
+DEBUG:PIL.Image:Importing CurImagePlugin
+DEBUG:PIL.Image:Importing DcxImagePlugin
+DEBUG:PIL.Image:Importing DdsImagePlugin
+DEBUG:PIL.Image:Importing EpsImagePlugin
+DEBUG:PIL.Image:Importing FitsImagePlugin
+DEBUG:PIL.Image:Importing FitsStubImagePlugin
+DEBUG:PIL.Image:Importing FliImagePlugin
+DEBUG:PIL.Image:Importing FpxImagePlugin
+DEBUG:PIL.Image:Image: failed to import FpxImagePlugin: No module named 'olefile'
+DEBUG:PIL.Image:Importing FtexImagePlugin
+DEBUG:PIL.Image:Importing GbrImagePlugin
+DEBUG:PIL.Image:Importing GifImagePlugin
+DEBUG:PIL.Image:Importing GribStubImagePlugin
+DEBUG:PIL.Image:Importing Hdf5StubImagePlugin
+DEBUG:PIL.Image:Importing IcnsImagePlugin
+DEBUG:PIL.Image:Importing IcoImagePlugin
+DEBUG:PIL.Image:Importing ImImagePlugin
+DEBUG:PIL.Image:Importing ImtImagePlugin
+DEBUG:PIL.Image:Importing IptcImagePlugin
+DEBUG:PIL.Image:Importing JpegImagePlugin
+DEBUG:PIL.Image:Importing Jpeg2KImagePlugin
+DEBUG:PIL.Image:Importing McIdasImagePlugin
+DEBUG:PIL.Image:Importing MicImagePlugin
+DEBUG:PIL.Image:Image: failed to import MicImagePlugin: No module named 'olefile'
+DEBUG:PIL.Image:Importing MpegImagePlugin
+DEBUG:PIL.Image:Importing MpoImagePlugin
+DEBUG:PIL.Image:Importing MspImagePlugin
+DEBUG:PIL.Image:Importing PalmImagePlugin
+DEBUG:PIL.Image:Importing PcdImagePlugin
+DEBUG:PIL.Image:Importing PcxImagePlugin
+DEBUG:PIL.Image:Importing PdfImagePlugin
+DEBUG:PIL.Image:Importing PixarImagePlugin
+DEBUG:PIL.Image:Importing PngImagePlugin
+DEBUG:PIL.Image:Importing PpmImagePlugin
+DEBUG:PIL.Image:Importing PsdImagePlugin
+DEBUG:PIL.Image:Importing QoiImagePlugin
+DEBUG:PIL.Image:Importing SgiImagePlugin
+DEBUG:PIL.Image:Importing SpiderImagePlugin
+DEBUG:PIL.Image:Importing SunImagePlugin
+DEBUG:PIL.Image:Importing TgaImagePlugin
+DEBUG:PIL.Image:Importing TiffImagePlugin
+DEBUG:PIL.Image:Importing WebPImagePlugin
+DEBUG:PIL.Image:Importing WmfImagePlugin
+DEBUG:PIL.Image:Importing XbmImagePlugin
+DEBUG:PIL.Image:Importing XpmImagePlugin
+DEBUG:PIL.Image:Importing XVThumbImagePlugin
+
+ 12%|█▎        | 1/8 [01:14<08:41, 74.51s/it]
+ 25%|██▌       | 2/8 [02:29<07:28, 74.79s/it]
+ 38%|███▊      | 3/8 [03:44<06:14, 74.81s/it]
+ 50%|█████     | 4/8 [04:59<04:59, 74.78s/it]
+ 62%|██████▎   | 5/8 [06:13<03:44, 74.73s/it]
+ 75%|███████▌  | 6/8 [07:28<02:29, 74.66s/it]
+ 88%|████████▊ | 7/8 [08:42<01:14, 74.56s/it]
+100%|██████████| 8/8 [09:56<00:00, 74.51s/it]
+100%|██████████| 8/8 [09:56<00:00, 74.62s/it]
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 1: generating actions ...
+>>> Step 1: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 2: generating actions ...
+>>> Step 2: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 3: generating actions ...
+>>> Step 3: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 4: generating actions ...
+>>> Step 4: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 5: generating actions ...
+>>> Step 5: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>

unitree_z1_dual_arm_cleanup_pencils/case1/psnr_result.json

@@ -0,0 +1,5 @@
+{
+    "gt_video": "unitree_z1_dual_arm_cleanup_pencils/case1/unitree_z1_dual_arm_cleanup_pencils_case1.mp4",
+    "pred_video": "unitree_z1_dual_arm_cleanup_pencils/case1/output/inference/0_full_fs4.mp4",
+    "psnr": 47.911564449209735
+}

unitree_z1_dual_arm_cleanup_pencils/case1/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_cleanup_pencils/case1"
+dataset="unitree_z1_dual_arm_cleanup_pencils"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_cleanup_pencils/case1/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 8 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_cleanup_pencils/case1/world_model_interaction_prompts/unitree_z1_dual_arm_cleanup_pencils.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+0,x,x,unitree_z1_dual_arm_cleanup_pencils,clean up eraser and pencils,x,x,x,Z1_Dual_Dex1,30

unitree_z1_dual_arm_cleanup_pencils/case2/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_cleanup_pencils/case2"
+dataset="unitree_z1_dual_arm_cleanup_pencils"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_cleanup_pencils/case2/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 8 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_cleanup_pencils/case2/world_model_interaction_prompts/unitree_z1_dual_arm_cleanup_pencils.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+50,x,x,unitree_z1_dual_arm_cleanup_pencils,clean up eraser and pencils,x,x,x,Z1_Dual_Dex1,30

unitree_z1_dual_arm_cleanup_pencils/case3/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_cleanup_pencils/case3"
+dataset="unitree_z1_dual_arm_cleanup_pencils"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_cleanup_pencils/case3/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 8 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_cleanup_pencils/case3/world_model_interaction_prompts/unitree_z1_dual_arm_cleanup_pencils.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+100,x,x,unitree_z1_dual_arm_cleanup_pencils,clean up eraser and pencils,x,x,x,Z1_Dual_Dex1,30

unitree_z1_dual_arm_cleanup_pencils/case4/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_cleanup_pencils/case4"
+dataset="unitree_z1_dual_arm_cleanup_pencils"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_cleanup_pencils/case4/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 8 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_cleanup_pencils/case4/world_model_interaction_prompts/unitree_z1_dual_arm_cleanup_pencils.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+200,x,x,unitree_z1_dual_arm_cleanup_pencils,clean up eraser and pencils,x,x,x,Z1_Dual_Dex1,30

unitree_z1_dual_arm_stackbox/case1/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_stackbox/case1"
+dataset="unitree_z1_dual_arm_stackbox"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_stackbox/case1/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 7 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_stackbox/case1/world_model_interaction_prompts/unitree_z1_dual_arm_stackbox.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+5,x,x,unitree_z1_dual_arm_stackbox,"Pick up the red cup on the table.",x,x,x,Unitree Z1 Robot Dual-Arm,30

unitree_z1_dual_arm_stackbox/case2/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_stackbox/case2"
+dataset="unitree_z1_dual_arm_stackbox"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_stackbox/case2/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 7 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_stackbox/case2/world_model_interaction_prompts/unitree_z1_dual_arm_stackbox.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+15,x,x,unitree_z1_dual_arm_stackbox,"Pick up the red cup on the table.",x,x,x,Unitree Z1 Robot Dual-Arm,30

unitree_z1_dual_arm_stackbox/case3/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_stackbox/case3"
+dataset="unitree_z1_dual_arm_stackbox"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_stackbox/case3/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 7 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_stackbox/case3/world_model_interaction_prompts/unitree_z1_dual_arm_stackbox.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+25,x,x,unitree_z1_dual_arm_stackbox,"Pick up the red cup on the table.",x,x,x,Unitree Z1 Robot Dual-Arm,30

unitree_z1_dual_arm_stackbox/case4/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_stackbox/case4"
+dataset="unitree_z1_dual_arm_stackbox"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_stackbox/case4/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 7 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_stackbox/case4/world_model_interaction_prompts/unitree_z1_dual_arm_stackbox.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+35,x,x,unitree_z1_dual_arm_stackbox,"Pick up the red cup on the table.",x,x,x,Unitree Z1 Robot Dual-Arm,30

unitree_z1_dual_arm_stackbox_v2/case1/output.log

@@ -0,0 +1,130 @@
+2026-02-10 17:39:22.590654: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
+2026-02-10 17:39:22.640645: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
+2026-02-10 17:39:22.640689: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
+2026-02-10 17:39:22.642010: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
+2026-02-10 17:39:22.649530: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
+To enable the following instructions: AVX2 AVX512F AVX512_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
+2026-02-10 17:39:23.575804: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
+Global seed set to 123
+INFO:mainlogger:LatentVisualDiffusion: Running in v-prediction mode
+INFO:unifolm_wma.models.diffusion_head.conditional_unet1d:number of parameters: 5.010531e+08
+INFO:unifolm_wma.models.diffusion_head.conditional_unet1d:number of parameters: 5.010531e+08
+AE working on z of shape (1, 4, 32, 32) = 4096 dimensions.
+INFO:root:Loaded ViT-H-14 model config.
+DEBUG:urllib3.connectionpool:Starting new HTTPS connection (1): hf-mirror.com:443
+DEBUG:urllib3.connectionpool:https://hf-mirror.com:443 "HEAD /laion/CLIP-ViT-H-14-laion2B-s32B-b79K/resolve/main/open_clip_pytorch_model.bin HTTP/1.1" 302 0
+INFO:root:Loading pretrained ViT-H-14 weights (laion2b_s32b_b79k).
+INFO:root:Loaded ViT-H-14 model config.
+DEBUG:urllib3.connectionpool:https://hf-mirror.com:443 "HEAD /laion/CLIP-ViT-H-14-laion2B-s32B-b79K/resolve/main/open_clip_pytorch_model.bin HTTP/1.1" 302 0
+INFO:root:Loading pretrained ViT-H-14 weights (laion2b_s32b_b79k).
+>>> model checkpoint loaded.
+>>> Load pre-trained model ...
+INFO:root:***** Configing Data *****
+>>> unitree_z1_stackbox: 1 data samples loaded.
+>>> unitree_z1_stackbox: data stats loaded.
+>>> unitree_z1_stackbox: normalizer initiated.
+>>> unitree_z1_dual_arm_stackbox: 1 data samples loaded.
+>>> unitree_z1_dual_arm_stackbox: data stats loaded.
+>>> unitree_z1_dual_arm_stackbox: normalizer initiated.
+>>> unitree_z1_dual_arm_stackbox_v2: 1 data samples loaded.
+>>> unitree_z1_dual_arm_stackbox_v2: data stats loaded.
+>>> unitree_z1_dual_arm_stackbox_v2: normalizer initiated.
+>>> unitree_z1_dual_arm_cleanup_pencils: 1 data samples loaded.
+>>> unitree_z1_dual_arm_cleanup_pencils: data stats loaded.
+>>> unitree_z1_dual_arm_cleanup_pencils: normalizer initiated.
+>>> unitree_g1_pack_camera: 1 data samples loaded.
+>>> unitree_g1_pack_camera: data stats loaded.
+>>> unitree_g1_pack_camera: normalizer initiated.
+>>> Dataset is successfully loaded ...
+>>> Generate 16 frames under each generation ...
+DEBUG:h5py._conv:Creating converter from 3 to 5
+DEBUG:PIL.PngImagePlugin:STREAM b'IHDR' 16 13
+DEBUG:PIL.PngImagePlugin:STREAM b'pHYs' 41 9
+DEBUG:PIL.PngImagePlugin:STREAM b'IDAT' 62 4096
+
+  0%|          | 0/11 [00:00<?, ?it/s]>>> Step 0: generating actions ...
+>>> Step 0: interacting with world model ...
+DEBUG:PIL.Image:Importing BlpImagePlugin
+DEBUG:PIL.Image:Importing BmpImagePlugin
+DEBUG:PIL.Image:Importing BufrStubImagePlugin
+DEBUG:PIL.Image:Importing CurImagePlugin
+DEBUG:PIL.Image:Importing DcxImagePlugin
+DEBUG:PIL.Image:Importing DdsImagePlugin
+DEBUG:PIL.Image:Importing EpsImagePlugin
+DEBUG:PIL.Image:Importing FitsImagePlugin
+DEBUG:PIL.Image:Importing FitsStubImagePlugin
+DEBUG:PIL.Image:Importing FliImagePlugin
+DEBUG:PIL.Image:Importing FpxImagePlugin
+DEBUG:PIL.Image:Image: failed to import FpxImagePlugin: No module named 'olefile'
+DEBUG:PIL.Image:Importing FtexImagePlugin
+DEBUG:PIL.Image:Importing GbrImagePlugin
+DEBUG:PIL.Image:Importing GifImagePlugin
+DEBUG:PIL.Image:Importing GribStubImagePlugin
+DEBUG:PIL.Image:Importing Hdf5StubImagePlugin
+DEBUG:PIL.Image:Importing IcnsImagePlugin
+DEBUG:PIL.Image:Importing IcoImagePlugin
+DEBUG:PIL.Image:Importing ImImagePlugin
+DEBUG:PIL.Image:Importing ImtImagePlugin
+DEBUG:PIL.Image:Importing IptcImagePlugin
+DEBUG:PIL.Image:Importing JpegImagePlugin
+DEBUG:PIL.Image:Importing Jpeg2KImagePlugin
+DEBUG:PIL.Image:Importing McIdasImagePlugin
+DEBUG:PIL.Image:Importing MicImagePlugin
+DEBUG:PIL.Image:Image: failed to import MicImagePlugin: No module named 'olefile'
+DEBUG:PIL.Image:Importing MpegImagePlugin
+DEBUG:PIL.Image:Importing MpoImagePlugin
+DEBUG:PIL.Image:Importing MspImagePlugin
+DEBUG:PIL.Image:Importing PalmImagePlugin
+DEBUG:PIL.Image:Importing PcdImagePlugin
+DEBUG:PIL.Image:Importing PcxImagePlugin
+DEBUG:PIL.Image:Importing PdfImagePlugin
+DEBUG:PIL.Image:Importing PixarImagePlugin
+DEBUG:PIL.Image:Importing PngImagePlugin
+DEBUG:PIL.Image:Importing PpmImagePlugin
+DEBUG:PIL.Image:Importing PsdImagePlugin
+DEBUG:PIL.Image:Importing QoiImagePlugin
+DEBUG:PIL.Image:Importing SgiImagePlugin
+DEBUG:PIL.Image:Importing SpiderImagePlugin
+DEBUG:PIL.Image:Importing SunImagePlugin
+DEBUG:PIL.Image:Importing TgaImagePlugin
+DEBUG:PIL.Image:Importing TiffImagePlugin
+DEBUG:PIL.Image:Importing WebPImagePlugin
+DEBUG:PIL.Image:Importing WmfImagePlugin
+DEBUG:PIL.Image:Importing XbmImagePlugin
+DEBUG:PIL.Image:Importing XpmImagePlugin
+DEBUG:PIL.Image:Importing XVThumbImagePlugin
+
+  9%|▉         | 1/11 [00:35<05:55, 35.52s/it]
+ 18%|█▊        | 2/11 [01:11<05:21, 35.73s/it]
+ 27%|██▋       | 3/11 [01:47<04:48, 36.04s/it]
+ 36%|███▋      | 4/11 [02:24<04:13, 36.19s/it]
+ 45%|████▌     | 5/11 [03:00<03:37, 36.25s/it]
+ 55%|█████▍    | 6/11 [03:36<03:00, 36.16s/it]
+ 64%|██████▎   | 7/11 [04:12<02:24, 36.09s/it]
+ 73%|███████▎  | 8/11 [04:48<01:48, 36.08s/it]
+ 82%|████████▏ | 9/11 [05:24<01:12, 36.06s/it]
+ 91%|█████████ | 10/11 [06:00<00:36, 36.07s/it]
+100%|██████████| 11/11 [06:36<00:00, 36.07s/it]
+100%|██████████| 11/11 [06:36<00:00, 36.07s/it]
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 1: generating actions ...
+>>> Step 1: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 2: generating actions ...
+>>> Step 2: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 3: generating actions ...
+>>> Step 3: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 4: generating actions ...
+>>> Step 4: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 5: generating actions ...
+>>> Step 5: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 6: generating actions ...
+>>> Step 6: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>
+>>> Step 7: generating actions ...
+>>> Step 7: interacting with world model ...
+>>>>>>>>>>>>>>>>>>>>>>>>

unitree_z1_dual_arm_stackbox_v2/case1/psnr_result.json

@@ -0,0 +1,5 @@
+{
+    "gt_video": "/home/qhy/unifolm-world-model-action/unitree_z1_dual_arm_stackbox_v2/case1/unitree_z1_dual_arm_stackbox_v2_case1.mp4",
+    "pred_video": "/home/qhy/unifolm-world-model-action/unitree_z1_dual_arm_stackbox_v2/case1/output/inference/5_full_fs4.mp4",
+    "psnr": 25.12008483689618
+}

unitree_z1_dual_arm_stackbox_v2/case1/run_world_model_interaction.sh

@@ -0,0 +1,25 @@
+res_dir="unitree_z1_dual_arm_stackbox_v2/case1"
+dataset="unitree_z1_dual_arm_stackbox_v2"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_stackbox_v2/case1/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 11 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae \
+        --fast_policy_no_decode
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_stackbox_v2/case1/world_model_interaction_prompts/unitree_z1_dual_arm_stackbox_v2.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+5,x,x,unitree_z1_dual_arm_stackbox_v2,"Stack the blocks in the rectangular block: red at the bottom, yellow in the middle, green on top",x,x,x,Unitree Z1 Robot Dual-Arm,30

unitree_z1_dual_arm_stackbox_v2/case2/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_stackbox_v2/case2"
+dataset="unitree_z1_dual_arm_stackbox_v2"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_stackbox_v2/case2/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 11 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_stackbox_v2/case2/world_model_interaction_prompts/unitree_z1_dual_arm_stackbox_v2.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+15,x,x,unitree_z1_dual_arm_stackbox_v2,"Stack the blocks in the rectangular block: red at the bottom, yellow in the middle, green on top",x,x,x,Unitree Z1 Robot Dual-Arm,30

unitree_z1_dual_arm_stackbox_v2/case3/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_stackbox_v2/case3"
+dataset="unitree_z1_dual_arm_stackbox_v2"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_stackbox_v2/case3/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 11 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"

unitree_z1_dual_arm_stackbox_v2/case3/world_model_interaction_prompts/unitree_z1_dual_arm_stackbox_v2.csv

@@ -0,0 +1,2 @@
+videoid,contentUrl,duration,data_dir,instruction,dynamic_confidence,dynamic_wording,dynamic_source_category,embodiment,fps
+25,x,x,unitree_z1_dual_arm_stackbox_v2,"Stack the blocks in the rectangular block: red at the bottom, yellow in the middle, green on top",x,x,x,Unitree Z1 Robot Dual-Arm,30

unitree_z1_dual_arm_stackbox_v2/case4/run_world_model_interaction.sh

@@ -0,0 +1,24 @@
+res_dir="unitree_z1_dual_arm_stackbox_v2/case4"
+dataset="unitree_z1_dual_arm_stackbox_v2"
+
+{
+    time CUDA_VISIBLE_DEVICES=0 python3 scripts/evaluation/world_model_interaction.py \
+        --seed 123 \
+        --ckpt_path ckpts/unifolm_wma_dual.ckpt \
+        --config configs/inference/world_model_interaction.yaml \
+        --savedir "${res_dir}/output" \
+        --bs 1 --height 320 --width 512 \
+        --unconditional_guidance_scale 1.0 \
+        --ddim_steps 50 \
+        --ddim_eta 1.0 \
+        --prompt_dir "unitree_z1_dual_arm_stackbox_v2/case4/world_model_interaction_prompts" \
+        --dataset ${dataset} \
+        --video_length 16 \
+        --frame_stride 4 \
+        --n_action_steps 16 \
+        --exe_steps 16 \
+        --n_iter 11 \
+        --timestep_spacing 'uniform_trailing' \
+        --guidance_rescale 0.7 \
+        --perframe_ae
+} 2>&1 | tee "${res_dir}/output.log"