Upload GameWorld interaction traces
Browse files- README.md +46 -0
- data/interactions.jsonl.gz +3 -0
- data/runs.jsonl.gz +3 -0
- metadata/export_manifest.json +0 -0
README.md
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
task_categories:
|
| 4 |
+
- visual-question-answering
|
| 5 |
+
- reinforcement-learning
|
| 6 |
+
language:
|
| 7 |
+
- en
|
| 8 |
+
tags:
|
| 9 |
+
- gameworld
|
| 10 |
+
- multimodal-agents
|
| 11 |
+
- interaction-traces
|
| 12 |
+
- browser-games
|
| 13 |
+
pretty_name: GameWorld Interaction Traces
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
# GameWorld Interaction Traces
|
| 17 |
+
|
| 18 |
+
This dataset contains model interaction traces exported from local GameWorld
|
| 19 |
+
evaluation runs. Each row in `data/interactions.jsonl.gz` represents one model
|
| 20 |
+
decision step, including the prompt, model response, parsed action, runtime
|
| 21 |
+
state, and task evaluation fields available in the original replay.
|
| 22 |
+
|
| 23 |
+
## Files
|
| 24 |
+
|
| 25 |
+
- `data/runs.jsonl.gz`: one row per rollout.
|
| 26 |
+
- `data/interactions.jsonl.gz`: one row per model interaction step.
|
| 27 |
+
- `metadata/export_manifest.json`: export counts and provenance.
|
| 28 |
+
|
| 29 |
+
## Export Summary
|
| 30 |
+
|
| 31 |
+
- Dataset repo target: `Fascetta/gameworld-interactions`
|
| 32 |
+
- Exported runs: 2067
|
| 33 |
+
- Exported interactions: 181447
|
| 34 |
+
- Exported at: 2026-09-18T18:20:10.060017+00:00
|
| 35 |
+
|
| 36 |
+
## Source
|
| 37 |
+
|
| 38 |
+
The traces were produced with the GameWorld benchmark runtime in this
|
| 39 |
+
repository. Raw local result directories are not required to load this dataset;
|
| 40 |
+
the source-relative paths are retained for provenance and auditability.
|
| 41 |
+
|
| 42 |
+
## Caveats
|
| 43 |
+
|
| 44 |
+
The dataset contains model prompts and model outputs from automated benchmark
|
| 45 |
+
runs. Screenshots are referenced by path when present in the replay metadata,
|
| 46 |
+
but this export does not copy bulky browser artifacts or local model weights.
|
data/interactions.jsonl.gz
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4a6cda0c1e5c49fff262dfbe93b18d4af74d11bde2b26310cec034dc8502a2cc
|
| 3 |
+
size 82263227
|
data/runs.jsonl.gz
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5cf8e9608f6b1b78b13686470c8d1d69bf648b7bd9e670ac3c1272fa30c179a6
|
| 3 |
+
size 174593
|
metadata/export_manifest.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|