Datasets:

lbernat
/

MAGMA-BENCH

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
schema_version: string
episode_id: string
scenario_id: string
skeleton_id: string
condition: string
interventions: list<item: null>
  child 0, item: null
stages: list<item: struct<kind: string, id: string, stage_type: string, arguments: struct<reception_objects: (... 1301 chars omitted)
  child 0, item: struct<kind: string, id: string, stage_type: string, arguments: struct<reception_objects: list<item: (... 1289 chars omitted)
      child 0, kind: string
      child 1, id: string
      child 2, stage_type: string
      child 3, arguments: struct<reception_objects: list<item: string>, minimum_completed_goals: int64, instruction: string, f (... 784 chars omitted)
          child 0, reception_objects: list<item: string>
              child 0, item: string
          child 1, minimum_completed_goals: int64
          child 2, instruction: string
          child 3, flag_answer: bool
          child 4, physical_assignment: list<item: struct<targets: list<item: string>, products: struct<electronics: int64, drinks: int64, s (... 47 chars omitted)
              child 0, item: struct<targets: list<item: string>, products: struct<electronics: int64, drinks: int64, snacks: int6 (... 35 chars omitted)
                  child 0, targets: list<item: string>
                      child 0, item: string
                  child 1, products: struct<electronics: int64, drinks: int64, snacks: int64, hygiene: int64, textile: int64>
                      child 0, electronics: int64
                   
...

              child 0, instruction: struct<type: string, arguments: struct<content: string, timestamp: int64, has_constraint: bool>>
                  child 0, type: string
                  child 1, arguments: struct<content: string, timestamp: int64, has_constraint: bool>
                      child 0, content: string
                      child 1, timestamp: int64
                      child 2, has_constraint: bool
              child 1, answer_after_completion: bool
              child 2, linked_to_previous: bool
          child 1, verification_prompt: string
          child 2, stage_goal_description: string
      child 6, reset_at_end: bool
      child 7, additive_stage: bool
      child 8, allow_tools_before_answer: bool
      child 9, allowed_tools: list<item: null>
          child 0, item: null
      child 10, active_errors: list<item: null>
          child 0, item: null
semantic_id: string
metadata: struct<clean_oracle_tool_calls: int64, episode_oracle_tool_calls: int64, length_bucket: string, numb (... 117 chars omitted)
  child 0, clean_oracle_tool_calls: int64
  child 1, episode_oracle_tool_calls: int64
  child 2, length_bucket: string
  child 3, number_of_stages: int64
  child 4, number_of_interventions: int64
  child 5, intervention_lags: list<item: null>
      child 0, item: null
  child 6, control_episode_id: string
description: string
environment_id: string
skill_types: list<item: null>
  child 0, item: null
name: string
definition: string
tools_type: string
to
{'schema_version': Value('string'), 'scenario_id': Value('string'), 'name': Value('string'), 'definition': Value('string'), 'environment_id': Value('string'), 'tools_type': Value('string'), 'skill_types': List(Value('null')), 'description': Value('string')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              schema_version: string
              episode_id: string
              scenario_id: string
              skeleton_id: string
              condition: string
              interventions: list<item: null>
                child 0, item: null
              stages: list<item: struct<kind: string, id: string, stage_type: string, arguments: struct<reception_objects: (... 1301 chars omitted)
                child 0, item: struct<kind: string, id: string, stage_type: string, arguments: struct<reception_objects: list<item: (... 1289 chars omitted)
                    child 0, kind: string
                    child 1, id: string
                    child 2, stage_type: string
                    child 3, arguments: struct<reception_objects: list<item: string>, minimum_completed_goals: int64, instruction: string, f (... 784 chars omitted)
                        child 0, reception_objects: list<item: string>
                            child 0, item: string
                        child 1, minimum_completed_goals: int64
                        child 2, instruction: string
                        child 3, flag_answer: bool
                        child 4, physical_assignment: list<item: struct<targets: list<item: string>, products: struct<electronics: int64, drinks: int64, s (... 47 chars omitted)
                            child 0, item: struct<targets: list<item: string>, products: struct<electronics: int64, drinks: int64, snacks: int6 (... 35 chars omitted)
                                child 0, targets: list<item: string>
                                    child 0, item: string
                                child 1, products: struct<electronics: int64, drinks: int64, snacks: int64, hygiene: int64, textile: int64>
                                    child 0, electronics: int64
                                 
              ...
              
                            child 0, instruction: struct<type: string, arguments: struct<content: string, timestamp: int64, has_constraint: bool>>
                                child 0, type: string
                                child 1, arguments: struct<content: string, timestamp: int64, has_constraint: bool>
                                    child 0, content: string
                                    child 1, timestamp: int64
                                    child 2, has_constraint: bool
                            child 1, answer_after_completion: bool
                            child 2, linked_to_previous: bool
                        child 1, verification_prompt: string
                        child 2, stage_goal_description: string
                    child 6, reset_at_end: bool
                    child 7, additive_stage: bool
                    child 8, allow_tools_before_answer: bool
                    child 9, allowed_tools: list<item: null>
                        child 0, item: null
                    child 10, active_errors: list<item: null>
                        child 0, item: null
              semantic_id: string
              metadata: struct<clean_oracle_tool_calls: int64, episode_oracle_tool_calls: int64, length_bucket: string, numb (... 117 chars omitted)
                child 0, clean_oracle_tool_calls: int64
                child 1, episode_oracle_tool_calls: int64
                child 2, length_bucket: string
                child 3, number_of_stages: int64
                child 4, number_of_interventions: int64
                child 5, intervention_lags: list<item: null>
                    child 0, item: null
                child 6, control_episode_id: string
              description: string
              environment_id: string
              skill_types: list<item: null>
                child 0, item: null
              name: string
              definition: string
              tools_type: string
              to
              {'schema_version': Value('string'), 'scenario_id': Value('string'), 'name': Value('string'), 'definition': Value('string'), 'environment_id': Value('string'), 'tools_type': Value('string'), 'skill_types': List(Value('null')), 'description': Value('string')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

MAGMA-BENCH 1.0.0

MAGMA-BENCH evaluates agents on interactive robotic tasks in simulation. This release contains 1,254 compiled episodes across 19 scenarios, with three semantic variations per task skeleton and clean, mission-update, interruption, execution-error, and combined conditions as defined by each skeleton.

These are executable benchmark artifacts loaded by magma_bench. The files include scenario definitions, initializations, stages, interventions, and evaluation specifications.

Installation

Use Python 3.12 and the exact package versions recorded in manifest.json:

python -m pip install "magma_bench==2.0.1" "magma_core[simulation]==2.0.1" "magma_scenarios==2.1.1"

Simulation requires compatible system drivers. Start a MAGMA protocol-v2 agent server exposing /health, /v1/info, and /v1/responses, and configure a verifier backend or use --skip-judge when appropriate.

Download and execution

magma-bench fetch --version 1.0.0
magma-bench run --dataset 1.0.0 \
  --results-path ./eval/experiment-1 \
  --agent-address http://127.0.0.1:8888 --videos off \
  --run-name experiment-1

magma-bench fetch without --version downloads the newest stable tagged dataset release. The default path is ${XDG_CACHE_HOME:-~/.cache}/magma-bench/datasets/v1.0.0/. Each version has its own directory; run --dataset uses downloaded files offline and checks the manifest's exact package pins before initializing the runtime.

For a specific destination:

magma-bench fetch --version 1.0.0 --path ./compiled-benchmark
magma-bench run --benchmark-root ./compiled-benchmark \
  --results-path ./eval/experiment-1 \
  --agent-address http://127.0.0.1:8888 --videos off

Core 2.0.1 parses its default YAML video setting as a boolean. Pass --videos off explicitly, as in these examples, or configure a quoted videos: "off" value in your benchmark configuration.

File organization

manifest.json                   Release version and exact package requirements
benchmark.json                  Complete episode index (artifact schema 1.8)
stats.json                      Compiled benchmark statistics
<scenario>/scenario.json        Scenario definition
<scenario>/skeleton_*/skeleton.json
<scenario>/skeleton_*/semantic_*/semantic.json
<scenario>/skeleton_*/semantic_*/episodes/*.json

Scenarios

advanced_delivery, bi_robot_sorting, coffee, coffee_comp, coffee_team, color_sorting, color_sorting_comp, delivery_box, hall_sorting, laundry, packaging, press_button, simp_advanced_delivery, simp_bi_robot_sorting, simp_hall_sorting, table_cleaning, table_cleaning_multi_robot, warehouse, and warehouse_cat.

Reproducibility and release policy

Dataset versions are independent of Python package versions. This release pins magma_bench==2.0.1, magma_core==2.0.1, and magma_scenarios==2.1.1.

License and links

The benchmark artifacts are distributed under BSD-2-Clause; see LICENSE.

Downloads last month: 1,346

Total file size:

39.4 MB

Collection including lbernat/MAGMA-BENCH

MAGMA

Collection

Ressources linked to MAGMA. My open source thesis framework to build your own robotic agent • 2 items • Updated 2 days ago