[WIP] robot arm experiment

[hengtaoguo]

Description

Tests

# Ckpt conversion
python -m maxtext.checkpoint_conversion.to_maxtext src/maxtext/configs/base.yml model_name=cosmos3-nano-reasoner base_output_directory=/dev/shm/hengtaoguo/checkpoints hf_access_token=xxx hardware=cpu skip_jax_distributed_system=True checkpoint_storage_use_ocdbt=False checkpoint_storage_use_zarr3=False use_multimodal=True scan_layers=false --eager_load_method=safetensors --lazy_load_tensors=False

# Decode
python -m maxtext.inference.decode src/maxtext/configs/base.yml model_name=cosmos3-nano-reasoner load_parameters_path=/dev/shm/hengtaoguo/checkpoints/0/items prompt=You\ are\ a\ robotic\ arm\ planner.\ The\ task\ is\ to\ put\ the\ flower\ into\ the\ red\ bottle.\ Generate\ a\ detailed\ plan\ consisting\ of\ sequential\ subtasks\ to\ accomplish\ the\ task. image_path=tests/assets/test_image_reasoning.jpg max_prefill_predict_length=1024 max_target_length=1124 hf_access_token=xxx scan_layers=false use_multimodal=true tokenizer_type=huggingface tokenizer_path=nvidia/Cosmos3-Nano ici_tensor_parallelism=4

# Result
Input `<|im_start|>user
<|vision_start|><|image_pad|><|vision_end|>You are a robotic arm planner. The task is to put the flower into the red bottle. Generate a detailed plan consisting of sequential subtasks to accomplish the task.<|im_end|>
<|im_start|>assistant
` -> `Move the arm to the flower. Grasp the flower. Move the arm to the red bottle. Place the flower in the red bottle.

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[codecov]

Codecov Report

❌ Patch coverage is 2.61438% with 149 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...xtext/checkpoint_conversion/utils/param_mapping.py	1.47%	134 Missing ⚠️
src/maxtext/multimodal/processor.py	0.00%	9 Missing and 1 partial ⚠️
src/maxtext/layers/encoders.py	0.00%	2 Missing ⚠️
src/maxtext/checkpoint_conversion/utils/utils.py	0.00%	1 Missing ⚠️
src/maxtext/layers/attentions.py	0.00%	0 Missing and 1 partial ⚠️
src/maxtext/layers/nnx_decoders.py	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!