pumpkin601 commited on
Commit
19a3b6e
·
verified ·
1 Parent(s): 4eba242

Link KineWorld source and mark checkpoint transfer in progress

Browse files
Files changed (1) hide show
  1. README.md +3 -1
README.md CHANGED
@@ -11,6 +11,8 @@ tags:
11
 
12
  # KineWorld step-500 checkpoint
13
 
 
 
14
  This repository contains the complete `step-500.safetensors` checkpoint from a KineWorld training run and its paired `action_norm_stats.npz`. The run completed 500 optimizer steps on 24 accelerators. This release includes a public training configuration with machine-specific paths removed.
15
 
16
  KineWorld uses a Wan2.2-TI2V-5B video backbone with an RGB/optical-flow dual-stream model and an action-prediction module. The run used RoboTwin head-camera data, 14-dimensional actions, and robot-only optical flow. It was initialized from an earlier action/world-model checkpoint rather than trained from scratch. The warm-start checkpoint SHA-256 is `21940bd45e95dc39bc3b166c6ecd2a777efa01c8068124958778173cd01ddf66`.
@@ -36,4 +38,4 @@ The checkpoint file is complete and its local SHA-256 was verified before upload
36
  - Warm-start checkpoint SHA-256: `21940bd45e95dc39bc3b166c6ecd2a777efa01c8068124958778173cd01ddf66`.
37
  - Hyperparameters and model modes: `training_config_public.json`.
38
 
39
- For source-code setup and inference commands, see the KineWorld GitHub repository linked from this model's Hugging Face page once published.
 
11
 
12
  # KineWorld step-500 checkpoint
13
 
14
+ **Release status:** the checkpoint file is uploading. The model card, public configuration, checksums, and paired action-normalization statistics are available.
15
+
16
  This repository contains the complete `step-500.safetensors` checkpoint from a KineWorld training run and its paired `action_norm_stats.npz`. The run completed 500 optimizer steps on 24 accelerators. This release includes a public training configuration with machine-specific paths removed.
17
 
18
  KineWorld uses a Wan2.2-TI2V-5B video backbone with an RGB/optical-flow dual-stream model and an action-prediction module. The run used RoboTwin head-camera data, 14-dimensional actions, and robot-only optical flow. It was initialized from an earlier action/world-model checkpoint rather than trained from scratch. The warm-start checkpoint SHA-256 is `21940bd45e95dc39bc3b166c6ecd2a777efa01c8068124958778173cd01ddf66`.
 
38
  - Warm-start checkpoint SHA-256: `21940bd45e95dc39bc3b166c6ecd2a777efa01c8068124958778173cd01ddf66`.
39
  - Hyperparameters and model modes: `training_config_public.json`.
40
 
41
+ For source-code setup and inference commands, see the [KineWorld GitHub repository](https://github.com/pumpkin601/KineWorld).