SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2502.13130 · CVPR · 2025

Magma: A Foundation Model for Multimodal AI Agents

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 16 functions out of this paper's own repositories and ran 12 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
microsoft/Magma canonical 12 of 16
FunctionStatusWhere it lives
IoU Ran microsoft/Magma/agents/ui_agent/util/box_annotator.py
code served (permissive licence) · get_code("9a3400bbcbc29023")
box_area Ran microsoft/Magma/agents/ui_agent/util/box_annotator.py
code served (permissive licence) · get_code("f631c3a13425e610")
draw_circle Ran microsoft/Magma/agents/robot_traj/utils/visualizer.py
code served (permissive licence) · get_code("7f5963ad8bf61e08")
draw_line Ran microsoft/Magma/agents/robot_traj/utils/visualizer.py
code served (permissive licence) · get_code("5caa64d42eda8676")
get_libero_obs Ran microsoft/Magma/agents/libero/libero_env_utils.py
code served (permissive licence) · get_code("bc05055a3a84f2c2")
get_magma_prompt Ran microsoft/Magma/agents/libero/libero_magma_utils.py
code served (permissive licence) · get_code("5e5dfc359305fb78")
intersection_area Ran microsoft/Magma/agents/ui_agent/util/box_annotator.py
code served (permissive licence) · get_code("23efd84e20b32497")
process_anyres_image Ran microsoft/Magma/magma/image_processing_magma.py
code served (permissive licence) · get_code("691aa4a9f8305f89")
process_anyres_image_global Ran microsoft/Magma/magma/image_processing_magma.py
code served (permissive licence) · get_code("d132cfaf9e0185f1")
resize_image Ran microsoft/Magma/agents/libero/libero_env_utils.py
code served (permissive licence) · get_code("090fa82a9a0f5b4b")
select_best_resolution Ran microsoft/Magma/magma/image_processing_magma.py
code served (permissive licence) · get_code("3999ff487573f32c")
split_to_even_chunks Ran microsoft/Magma/trainer/trainer.py
code served (permissive licence) · get_code("10893c4608c08075")
get_magma_action Not yet run microsoft/Magma/agents/libero/libero_magma_utils.py
code served (permissive licence) · get_code("d31031274c15dc34")
get_mm_adapter_state_maybe_zero_3 Not yet run microsoft/Magma/trainer/trainer.py
code served (permissive licence) · get_code("bb35e3ac741bb2c9")
maybe_zero_3 Not yet run microsoft/Magma/trainer/trainer.py
code served (permissive licence) · get_code("735025744c1ab0cf")
read_video_from_path Not yet run microsoft/Magma/agents/robot_traj/utils/visualizer.py
code served (permissive licence) · get_code("574c10b6f4d920f3")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Figure 1. We introduce Magma, the first foundation model that is capable of interpreting and grounding multimodal inputs, and taking actions towards a goal in both digital and physical environments. With our newly proposed pretraining techniques, Magma learns effectively from images, videos and robotics data to bridge verbal and spatial intelligence, taking a step further to an intelligent multimodal AI agent.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2502.13130")
get_code_for_paper("2502.13130")
have("2502.13130")

Connect an agent — have() is free.