Dataset Viewer
Auto-converted to Parquet Duplicate
num_samples
int64
hidden_size
int64
model_name
string
model_revision
string
max_model_len
int64
has_hard_negatives
bool
samples
list
123,849
4,096
Qwen/Qwen3-8B
b968826d9c46dd6066d109eabc6255188de91218
8,192
false
[{"task_id":"terminal-bench/uefi-bootkit","trajectory_id":"02a8f8f7-4d2b-4c06-afe7-f1551a8e3781","st(...TRUNCATED)

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Terminal trajectory embedding task subset

20 selected tasks, 1,281 trajectories, and 123,849 state/action pairs. Only trajectories with numeric reward == 1 are retained.

The saved vectors are exact selected rows of the existing embeddings; the encoder was not rerun. train/metadata.json matches both tensor row orders. source_row_indices.json records the original row indices; selection.json records selection parameters, source checksums and output checksums.

selected_tasks.json identifies the selected tasks. heldout_tasks.json lists the 46 other tasks, whose vectors remain in the full source. If this subset is used for training, evaluate on held-out tasks, excluding these selected tasks. A trajectory-level success label is not an action correctness label. All source context limits and truncation metadata remain unchanged.

Source embedding dataset: Hkang/terminal-bench-4-qwen3-8b-embeddings, commit 796ecf6ad08e4deaa06507fb5418ca60a74489ba.

This is a reproducible example using seed 42 and successful trajectories only; the full 66-task source remains available. It uses the original Qwen3-8B 8K embeddings with their existing truncation metadata.

Downloads last month
-