Hugging Face's logo

Datasets:
comfortably-dumb
/
dedeucebench_hud_dev

License:
Dataset card Data Studio Files
xet
Community
Dataset Preview
Duplicate
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code:   DatasetGenerationError
Exception:    ArrowNotImplementedError
Message:      Cannot write struct type 'arguments' with no child field to Parquet. Consider adding a dummy child field.
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/builder.py", line 1831, in _prepare_split_single
                  writer.write_table(table)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/arrow_writer.py", line 642, in write_table
                  self._build_writer(inferred_schema=pa_table.schema)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/arrow_writer.py", line 457, in _build_writer
                  self.pa_writer = self._WRITER_CLASS(self.stream, schema)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/pyarrow/parquet/core.py", line 1010, in __init__
                  self.writer = _parquet.ParquetWriter(
                File "pyarrow/_parquet.pyx", line 2157, in pyarrow._parquet.ParquetWriter.__cinit__
                File "pyarrow/error.pxi", line 154, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 91, in pyarrow.lib.check_status
              pyarrow.lib.ArrowNotImplementedError: Cannot write struct type 'arguments' with no child field to Parquet. Consider adding a dummy child field.
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/builder.py", line 1847, in _prepare_split_single
                  num_examples, num_bytes = writer.finalize()
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/arrow_writer.py", line 661, in finalize
                  self._build_writer(self.schema)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/arrow_writer.py", line 457, in _build_writer
                  self.pa_writer = self._WRITER_CLASS(self.stream, schema)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/pyarrow/parquet/core.py", line 1010, in __init__
                  self.writer = _parquet.ParquetWriter(
                File "pyarrow/_parquet.pyx", line 2157, in pyarrow._parquet.ParquetWriter.__cinit__
                File "pyarrow/error.pxi", line 154, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 91, in pyarrow.lib.check_status
              pyarrow.lib.ArrowNotImplementedError: Cannot write struct type 'arguments' with no child field to Parquet. Consider adding a dummy child field.
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1456, in compute_config_parquet_and_info_response
                  parquet_operations = convert_to_parquet(builder)
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1055, in convert_to_parquet
                  builder.download_and_prepare(
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/builder.py", line 894, in download_and_prepare
                  self._download_and_prepare(
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/builder.py", line 970, in _download_and_prepare
                  self._prepare_split(split_generator, **prepare_split_kwargs)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/builder.py", line 1702, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/builder.py", line 1858, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

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.

id
string
prompt
string
mcp_config
dict
setup_tool
dict
evaluate_tool
dict
agent_config
dict
allowed_tools
list
max_steps
int64
seed_2
OBSERVATION: {"alphabet": ["A", "B", "C"], "outputs": ["x", "y", "z"], "budget": 40, "goal": "basic", "n_states": 3, "target_len": 8, "trap": false} Episode notes: - alphabet: allowed input symbols ('A','B','C'). - outputs: each act(symbol) produces exactly one output in {'x','y','z'}. - budget: total act() calls available this episode. - n_states: total number of states (indices 0..n-1). In explanations we refer to them as s0, s1, ..., s{n-1}. Start is s0 (index 0). - trap: whether trap transitions may exist; if trap_hit becomes true, success is impossible. This is a single-trajectory, stateful, non-resetting episode: each act(symbol) advances the live machine and produces an output. Both output and next state depend on (current_state, symbol). There is no undo or rewind machine is always live. submit_table(...) does not reset state. Task: Use act() to gather enough evidence (observing the outputs), then call submit_table(table_json) with a complete table for all states s0..s{n-1} (indices 0..n-1 in JSON) and symbols A,B,C. If your table is incorrect, submit_table may return a counterexample (a short input sequence from the start state with the true outputs); you may use it to adjust your hypothesis. This consumes 1 query and the episode continues. Use only act() and submit_table(...). Always terminate by calling submit_table(...). Tool call schemas (short): - act(symbol: 'A'|'B'|'C') -> returns JSON {output:'x'|'y'|'z', budget_left:int, t:int, trap_hit:bool, queries_used:int}. - submit_table(table_json: string-of-JSON). The table_json MUST be an object with keys 'n', 'start', 'trans', and each symbol entry is an array [next_state:int, output:'x'|'y'|'z'] (do NOT use objects like {output:..., next_state:...}). Minimal example: {"n":2, "start":0, "trans":{"0":{"A":[1,"x"],"B":[1,"z"],"C":[1,"z"]},"1":{"A":[0,"z"],"B":[1,"z"],"C":[1,"z"]}}}.
{ "hud": { "headers": { "Authorization": "Bearer ${HUD_API_KEY}", "Mcp-Image": "docker.io/vedanshsharma123/dedeucebench_hud@sha256:4cd3bc40218f472d7ad945bd7a4d4aa38e3e0a77a775e92b5ac225d10042c3ee", "Run-Id": "${RUN_ID}" }, "url": "https://mcp.hud.so/v3/mcp" } }
{ "arguments": { "budget": 40, "feedback": true, "max_steps": 64, "mode": "basic", "n_states": 3, "seed": 2, "target_len": 8, "trap": false, "variety": false }, "name": "setup" }
{ "arguments": {}, "name": "evaluate" }
{ "system_prompt": "You are an autonomous tool-using agent interacting with a hidden Mealy machine (finite-state transducer).\nObjective: exactly identify the machine and submit the full transition table via submit_table(table_json).\nReturn ONLY function tool calls; never output natural language. All responses must be valid JSON if any content is emitted.\n\nBenchmark focus: identification-first. Success is achieved only by exact transition-table submission via submit_table(table_json).\n\nEpisode semantics:\n- Stateful episode: the hidden machine's state persists across all tool calls; there are no resets between calls.\n- act(symbol) produces exactly one output (a symbol in {'x','y','z'}) and advances the hidden state; both output and next state are deterministic functions of (current_state, symbol).\n- submit_table(table_json) does not change or reset state; wrong submissions consume 1 query and the episode continues; correct submission ends the episode.\n- Start state is 0; the hidden state updates only when you call act(symbol).\n- Each act() consumes 1 query from the budget (invalid symbols still consume 1 and return an error).\n- submit_table(table_json): if your table is incorrect, it consumes 1 query and does NOT end the episode (when feedback is enabled, a short counterexample is returned). If correct, it ends the episode and does not consume budget. When budget reaches 0, the episode ends with ok=false.\n\nTools (use only act() and submit_table()):\n- act(symbol: 'A'|'B'|'C') -> JSON {output, budget_left, t, trap_hit, queries_used}. Each call consumes 1 query, produces an output, and advances the hidden state.\n- submit_table(...) -> JSON {ok, budget_left, queries_used, trap_hit, counterexample?}. If ok=false, consumes 1 query and does NOT end the episode (counterexample present only when feedback is enabled). If ok=true, ends the episode.\n\nTool return fields (definitions):\n- output: one of {'x','y','z'} produced by the latest act() call.\n- budget_left: remaining number of act() queries.\n- t: 1-based step index since the episode started (increments on each act()).\n- trap_hit: boolean; once true it remains true for the rest of the episode.\n- queries_used: total count of act() calls so far.\n\nCounterexample semantics (when feedback is enabled):\n- If submit_table is incorrect, the environment may return a short distinguishing test starting from the start state (state 0): a sequence of inputs with the corresponding true outputs.\n- This counterexample is diagnostic only. It does NOT change the live episode state, and it is NOT tied to your current state trajectory.\n- You may use it to refine your hypothesis; then continue probing with act() and resubmit.\n\nSubmit-table JSON schema (table_json string must parse to this shape, strictly follow this):\nImportant: Each entry is [next_state:int, output:'x'|'y'|'z'] — do NOT swap to [output, next_state].\n\n{\n \"n\": <int total_states>,\n \"start\": 0,\n \"trans\": {\n \"0\": { \"A\": [<ns:int>, <output:\"x\"|\"y\"|\"z\">], \"B\": [<ns>, <output>], \"C\": [<ns>, <output>] },\n \"1\": { \"A\": [<ns>, <output>], \"B\": [<ns>, <output>], \"C\": [<ns>, <output>] },\n ... up to \"n-1\"\n }\n}\n\nSkeleton example of table_json (Strictly follow this) (for n=2 — adjust values):\n{\"n\":2,\"start\":0,\"trans\":{\"0\":{\"A\":[1,\"y\"],\"B\":[0,\"x\"],\"C\":[0,\"x\"]},\"1\":{\"A\":[0,\"x\"],\"B\":[1,\"y\"],\"C\":[1,\"z\"]}}}\n\nFormatting & compliance:\n- Respond only with function tool calls as per the provided tool schemas.\n- The submit_table argument must be a single JSON string (not an object) matching the schema mentioned. Note that it include n, start and trans parts.\n- Do NOT echo the observation or tool descriptions.\n- Ensure \"trans\" covers every state index 0..n-1 and each of A,B,C exactly once.\n- Always terminate by calling submit_table(...).", "temperature": 0, "top_p": 1 }
[ "act", "submit_table" ]
64
seed_2
OBSERVATION: {"alphabet": ["A", "B", "C"], "outputs": ["x", "y", "z"], "budget": 35, "goal": "basic", "n_states": 2, "target_len": 8, "trap": false} Episode notes: - alphabet: allowed input symbols ('A','B','C'). - outputs: each act(symbol) produces exactly one output in {'x','y','z'}. - budget: total act() calls available this episode. - n_states: total number of states (indices 0..n-1). In explanations we refer to them as s0, s1, ..., s{n-1}. Start is s0 (index 0). - trap: whether trap transitions may exist; if trap_hit becomes true, success is impossible. This is a single-trajectory, stateful, non-resetting episode: each act(symbol) advances the live machine and produces an output. Both output and next state depend on (current_state, symbol). There is no undo or rewind machine is always live. submit_table(...) does not reset state. Task: Use act() to gather enough evidence (observing the outputs), then call submit_table(table_json) with a complete table for all states s0..s{n-1} (indices 0..n-1 in JSON) and symbols A,B,C. If your table is incorrect, submit_table may return a counterexample (a short input sequence from the start state with the true outputs); you may use it to adjust your hypothesis. This consumes 1 query and the episode continues. Use only act() and submit_table(...). Always terminate by calling submit_table(...). Tool call schemas (short): - act(symbol: 'A'|'B'|'C') -> returns JSON {output:'x'|'y'|'z', budget_left:int, t:int, trap_hit:bool, queries_used:int}. - submit_table(table_json: string-of-JSON). The table_json MUST be an object with keys 'n', 'start', 'trans', and each symbol entry is an array [next_state:int, output:'x'|'y'|'z'] (do NOT use objects like {output:..., next_state:...}). Minimal example: {"n":2, "start":0, "trans":{"0":{"A":[1,"x"],"B":[1,"z"],"C":[1,"z"]},"1":{"A":[0,"z"],"B":[1,"z"],"C":[1,"z"]}}}.
{ "hud": { "headers": { "Authorization": "Bearer ${HUD_API_KEY}", "Mcp-Image": "docker.io/vedanshsharma123/dedeucebench_hud@sha256:4cd3bc40218f472d7ad945bd7a4d4aa38e3e0a77a775e92b5ac225d10042c3ee", "Run-Id": "${RUN_ID}" }, "url": "https://mcp.hud.so/v3/mcp" } }
{ "arguments": { "budget": 55, "feedback": true, "max_steps": 64, "mode": "basic", "n_states": 2, "seed": 2, "target_len": 8, "trap": false, "variety": false }, "name": "setup" }
{ "arguments": {}, "name": "evaluate" }
{ "system_prompt": "You are an autonomous tool-using agent interacting with a hidden Mealy machine (finite-state transducer).\nObjective: exactly identify the machine and submit the full transition table via submit_table(table_json).\nReturn ONLY function tool calls; never output natural language. All responses must be valid JSON if any content is emitted.\n\nBenchmark focus: identification-first. Success is achieved only by exact transition-table submission via submit_table(table_json).\n\nEpisode semantics:\n- Stateful episode: the hidden machine's state persists across all tool calls; there are no resets between calls.\n- act(symbol) produces exactly one output (a symbol in {'x','y','z'}) and advances the hidden state; both output and next state are deterministic functions of (current_state, symbol).\n- submit_table(table_json) does not change or reset state; wrong submissions consume 1 query and the episode continues; correct submission ends the episode.\n- Start state is 0; the hidden state updates only when you call act(symbol).\n- Each act() consumes 1 query from the budget (invalid symbols still consume 1 and return an error).\n- submit_table(table_json): if your table is incorrect, it consumes 1 query and does NOT end the episode (when feedback is enabled, a short counterexample is returned). If correct, it ends the episode and does not consume budget. When budget reaches 0, the episode ends with ok=false.\n\nTools (use only act() and submit_table()):\n- act(symbol: 'A'|'B'|'C') -> JSON {output, budget_left, t, trap_hit, queries_used}. Each call consumes 1 query, produces an output, and advances the hidden state.\n- submit_table(...) -> JSON {ok, budget_left, queries_used, trap_hit, counterexample?}. If ok=false, consumes 1 query and does NOT end the episode (counterexample present only when feedback is enabled). If ok=true, ends the episode.\n\nTool return fields (definitions):\n- output: one of {'x','y','z'} produced by the latest act() call.\n- budget_left: remaining number of act() queries.\n- t: 1-based step index since the episode started (increments on each act()).\n- trap_hit: boolean; once true it remains true for the rest of the episode.\n- queries_used: total count of act() calls so far.\n\nCounterexample semantics (when feedback is enabled):\n- If submit_table is incorrect, the environment may return a short distinguishing test starting from the start state (state 0): a sequence of inputs with the corresponding true outputs.\n- This counterexample is diagnostic only. It does NOT change the live episode state, and it is NOT tied to your current state trajectory.\n- You may use it to refine your hypothesis; then continue probing with act() and resubmit.\n\nSubmit-table JSON schema (table_json string must parse to this shape, strictly follow this):\nImportant: Each entry is [next_state:int, output:'x'|'y'|'z'] — do NOT swap to [output, next_state].\n\n{\n \"n\": <int total_states>,\n \"start\": 0,\n \"trans\": {\n \"0\": { \"A\": [<ns:int>, <output:\"x\"|\"y\"|\"z\">], \"B\": [<ns>, <output>], \"C\": [<ns>, <output>] },\n \"1\": { \"A\": [<ns>, <output>], \"B\": [<ns>, <output>], \"C\": [<ns>, <output>] },\n ... up to \"n-1\"\n }\n}\n\nSkeleton example of table_json (Strictly follow this) (for n=2 — adjust values):\n{\"n\":2,\"start\":0,\"trans\":{\"0\":{\"A\":[1,\"y\"],\"B\":[0,\"x\"],\"C\":[0,\"x\"]},\"1\":{\"A\":[0,\"x\"],\"B\":[1,\"y\"],\"C\":[1,\"z\"]}}}\n\nFormatting & compliance:\n- Respond only with function tool calls as per the provided tool schemas.\n- The submit_table argument must be a single JSON string (not an object) matching the schema mentioned. Note that it include n, start and trans parts.\n- Do NOT echo the observation or tool descriptions.\n- Ensure \"trans\" covers every state index 0..n-1 and each of A,B,C exactly once.\n- Always terminate by calling submit_table(...).", "temperature": 0, "top_p": 1 }
[ "act", "submit_table" ]
64
README.md exists but content is empty.
Downloads last month
6