lerobot/cosmos3-nano
Overview
Source roundup from published reports. Claims below are attributed to their publishers, not independently verified. LeRobot — Model repository updates: LeRobot Releases Cosmos3-Nano Policy Conversion. LeRobot has released a policy conversion of the NVIDIA Cosmos3-Nano base checkpoint, providing model components for robot policy post-training. The conversion does not include task-specific fine-tuning and requires a tar…
Generated from attributed reports · Updated 57 minutes ago
Event evidence and corrections
1 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · Stored scalar weights: 16454 other · Basis not reported
“mponents needed by the LeRobot policy interface. Item Value Original model developer NVIDIA Original checkpoint nvidia/Cosmos3-Nano Exact source revision e59a53c25979a090fa8706c9acc0c254a6e89b92 LeRobot compatibility baseline b8f060837ced0a4ce4b136ea65b4d56bc7894bed Policy type cosmos3_nano Backbone Qwen3-VL MoT Decoder layers / hidden width 36 / 4096 Stored scalar weights, including frozen components and VAE 16,454,”
Exact source · revision 2huggingface
Reported quantity · Number of decoder layers: 36 other · Basis not reported
“mponents needed by the LeRobot policy interface. Item Value Original model developer NVIDIA Original checkpoint nvidia/Cosmos3-Nano Exact source revision e59a53c25979a090fa8706c9acc0c254a6e89b92 LeRobot compatibility baseline b8f060837ced0a4ce4b136ea65b4d56bc7894bed Policy type cosmos3_nano Backbone Qwen3-VL MoT Decoder layers / hidden width 36 / 4096 Stored scalar weights, including frozen components and VAE 16,454,”
Exact source · revision 2huggingface
Reported quantity · Hidden width: 4096 other · Basis not reported
“mponents needed by the LeRobot policy interface. Item Value Original model developer NVIDIA Original checkpoint nvidia/Cosmos3-Nano Exact source revision e59a53c25979a090fa8706c9acc0c254a6e89b92 LeRobot compatibility baseline b8f060837ced0a4ce4b136ea65b4d56bc7894bed Policy type cosmos3_nano Backbone Qwen3-VL MoT Decoder layers / hidden width 36 / 4096 Stored scalar weights, including frozen components and VAE 16,454,”
Exact source · revision 2huggingface
Reported quantity · Number of time-embedder tensors: 4 other · Basis not reported
“n modules. The saved dtype=float32 supports FP32 master weights during training. Most stored weights retain upstream BF16 precision. The four time-embedder tensors are FP32. For inference set --policy.dtype=bfloat16 . For evaluation also set --tf32=false . Scope and evaluation The LeRobot implementation supports joint video-and-action wam mode, a single observation step, and the UniPC sampler. Actual action width com”
Exact source · revision 2huggingface
Evidence dependency changes (1)
- 2026-10-30T17:15:31.375Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
Report timeline
Follow attributed reports and material updates.
- LeRobot — Model repository updatesSignalLeRobot Releases Cosmos3-Nano Policy Conversion
LeRobot has released a policy conversion of the NVIDIA Cosmos3-Nano base checkpoint, providing model components for robot policy post-training. The conversion does not include task-specific fine-tuning and requires a target dataset to define frame rate and camera size.
Event coverage history
There is not enough continuous observation data to show a trend.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).