Datasets:

darwinagent
/

WIA-Bench

system stringclasses 211 values	prompt stringlengths 245 106k	ground_truth stringlengths 34 30.3k	domain stringclasses 7 values
Role: Operate as a language world model for a software-engineering shell testbed. You model how the environment changes after an issued action and report only the resulting observation. Input: Interpret the goal or task metadata, latest observation, action block, and earlier turns as a persistent state trace. The curr...	Turn 1 User ###[Task Description] Crash when unpacking non TypeVarTuple in args list to `Callable` alias <!-- Use this form only if mypy reports an "INTERNAL ERROR" and/or gives a traceback. Please include the traceback and all other messages below (use `mypy --show-traceback`). --> Crash Report ...	###[Observation] {"status": "SUCCEEDED", "exit_code": 0, "stdout": " if isinstance(repl, TypeVarTupleType):\n return repl\n raise NotImplementedError\n\n def visit_unpack_type(self, t: UnpackType) -> Type:\n # It is impossible to reasonably implement visit_unpack_type, because\n ...	swe
Role: You are a language world model for a web search and page-browsing environment. Your responsibility is to simulate the environment after an action has already been issued, not to act as the agent. Input: This prompt accepts the goal or task context, the latest `###[Observation]`, the current `###[Action]`, and an...	Turn 1 User ###[Goal] Find information about what WNZK is and the city where it is located. WNZK is likely a radio station or other type of location identifier. ###[Action] Use `browser.search` to search the web or retrieval index for relevant documents. - This advances the task or changes state. - Args: `query`="W...	###[Observation] Doc 220 URL: https://radio-locator.com/info/WNZK-AM L1: WNZK-AM 680 kHz L2: L3: Dearborn Heights, Michigan L4: "Station of the Nations" L5: L6: L7: Website: L8: https://www.birach.com/wnzk.html L9: L10: Audio Feed: L11: http://wnzk.birach.com:9000/stream L12: L13: Station Owner: L14: Birach Broadc...	search
Role: Act as a language world model for a web search and page-browsing environment. You predict the post-action observation of the environment; you are not the acting agent, planner, or tool executor. Input: Use the goal or task description, the most recent `###[Observation]`, the current `###[Action]`, and prior obse...	Turn 1 User ###[Goal] How many counties are there in the largest state in the region where the film The Family Stone was set? You should follow the format instruction in the requestion strictly and wrap the final answer in \boxed{}. ###[Action] Use `browser.search` to search the web or retrieval index for relevant ...	###[Observation] https://en.wikipedia.org/wiki/The_Family_Stone URL: https://en.wikipedia.org/wiki/The_Family_Stone L1: Document not found	search
Role: Act as a language world model for a terminal command-execution environment. You predict the post-action observation of the environment; you are not the acting agent, planner, or tool executor. Input: Use the goal or task description, the most recent `###[Observation]`, the current `###[Action]`, and prior observ...	Turn 1 User ###[Goal] Please solve this task: You are a platform engineer responsible for maintaining our CI/CD pipelines. We are building a new artifact validation system. There are two components to this system: 1. A small Rust CLI tool used by build agents to compute Adler-32 checksums of artifacts. 2. A Pyth...	###[Observation] [package] name = "rust_helper" version = "0.1.0" edition = "2021" [dependencies] adler = "1.0.2" (exit_code=0)	terminal
Role: You are a language world model for a desktop accessibility-control environment. Your responsibility is to simulate the environment after an action has already been issued, not to act as the agent. Input: This prompt accepts the goal or task context, the latest `###[Observation]`, the current `###[Action]`, and a...	Turn 1 User ###[Observation] Computer 'Program Manager', b=0,0,1920,1080, raw_control='PaneControl' node=1 generic '', raw_control='PaneControl', b=0,0,1920,1080 node=2 list 'Desktop', raw_control='ListControl', automation_id='1', b=0,0,1920,1080 node=3 listitem 'Recycle Bin', raw_control='ListItemControl', b...	###[Observation] Computer 'New Tab - Google Chrome', b=15,15,1590,993, raw_control='WindowControl' node=1 generic '', raw_control='PaneControl', b=26,15,1579,982 node=2 generic 'New Tab - Google Chrome', raw_control='PaneControl', b=25,15,1580,983 node=3 generic '', raw_control='PaneControl', b=25,15,1580,983 no...	os
Role: Act as a language world model for a browser accessibility environment with reference IDs. You predict the post-action observation of the environment; you are not the acting agent, planner, or tool executor. Input: Use the goal or task description, the most recent `###[Observation]`, the current `###[Action]`, an...	Turn 1 User ###[Observation] - RootWebArea "(2 unread) - webtasks@yahoo.com - Yahoo Mail" [ref=e1] - banner [ref=e2] - banner [ref=e3] - navigation [ref=e4] - list [ref=e5] - listitem [ref=e6] - link " HOME" [ref=e7] - graphics-symbol [ref=e8] - list...	###[Observation] - RootWebArea "(2 unread) - webtasks@yahoo.com - Yahoo Mail" [ref=e1] - banner [ref=e2] - banner [ref=e3] - navigation [ref=e4] - list [ref=e5] - listitem [ref=e6] - link " HOME" [ref=e7] - graphics-symbol [ref=e8] - listitem [ref=e9] ...	web
Role: Act as a language world model for a browser accessibility environment with reference IDs. You predict the post-action observation of the environment; you are not the acting agent, planner, or tool executor. Input: Use the goal or task description, the most recent `###[Observation]`, the current `###[Action]`, an...	Turn 1 User ###[Observation] - RootWebArea "(2 unread) - webtasks@yahoo.com - Yahoo Mail" [ref=e1] - banner [ref=e2] - banner [ref=e3] - navigation [ref=e4] - list [ref=e5] - listitem [ref=e6] - link " HOME" [ref=e7] - graphics-symbol [ref=e8] - list...	###[Observation] - RootWebArea "(2 unread) - webtasks@yahoo.com - Yahoo Mail" [ref=e1] - banner [ref=e2] - banner [ref=e3] - navigation [ref=e4] - list [ref=e5] - listitem [ref=e6] - link " HOME" [ref=e7] - graphics-symbol [ref=e8] - listitem [ref=e9] ...	web
"Role: You are a state-transition world model for a web search and page-browsing environment. Your p(...TRUNCATED)	"Turn 1 User\n###[Goal]\nRdio was a music streaming service that held apps on several platforms,(...TRUNCATED)	"###[Observation]\nhttps://en.m.wikipedia.org/wiki/Rdio\nURL: https://en.m.wikipedia.org/wiki/Rdio\n(...TRUNCATED)	search
"Role: You are an observation generator for a language world model of a web search and page-browsing(...TRUNCATED)	"Turn 1 User\n###[Goal]\nIn 2018, who was elected mayor of the place Sure of You takes place in?(...TRUNCATED)	"###[Observation]\nDoc 1997799\nURL: http://womensfictionwriters.com/2016/02/18/guest-post-author-yo(...TRUNCATED)	search
"Role: You are a state-transition world model for an MCP business-tool environment. Your prediction (...TRUNCATED)	"Turn 1 User\n###[Goal]\nRegister a new pet for existing owner 'Maria Sanchez' named 'Luna' with(...TRUNCATED)	"###[Observation]\n{\n \"id\": 8,\n \"owner_id\": 1,\n \"name\": \"Luna\",\n \"species\": \"Cani(...TRUNCATED)	mcp

End of preview. Expand in Data Studio

WIA-Bench

Files and Sample Counts

File	Purpose	Samples
`ID-Bench.jsonl`	In-distribution (ID) evaluation data	1,808
`OOD-Bench.jsonl`	Out-of-distribution (OOD) evaluation data	1,739

Field Definitions

Each sample contains four fields:

Field	Type	Purpose
`system`	string	System instruction or task rules.
`prompt`	string	Input context, task description, or history.
`ground_truth`	string	Target output or correct result.
`domain`	string	Task domain.

Domain Statistics

Domain	`ID-Bench.jsonl`	`OOD-Bench.jsonl`
`android`	157	254
`mcp`	209	121
`os`	295	221
`search`	293	297
`swe`	294	292
`terminal`	278	283
`web`	282	271
Total	1,808	1,739

Download

pip install -U huggingface_hub

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="LightningCreeper/WIA-Bench",
    repo_type="dataset",
    allow_patterns=["*.jsonl"],
    local_dir="WIA-Bench",
)

Downloads last month: 30

Collection including darwinagent/WIA-Bench

WIA Datasets

Collection

World-model datasets for the WIA project. • 4 items • Updated 10 days ago • 1