Datasetpublic

whywhywhyyy/qwen36-35b-a3b-clawbench

Qwen3.6-35B-A3B ClawBench Full Results This repository contains organized ClawBench run artifacts from bench/ClawBench/test-output/qwen36-35b-a3b-full-20260510-2140. The formal result directories are preserved under results/. Quarantined infrastructure-failure directories and top-level runtime control files such as .pid, .proc, .logpath, and .monitor-hermes-state/ are intentionally excluded from this organized export. Contents… See the full description on the dataset page: https://huggingface.co/datasets/whywhywhyyy/qwen36-35b-a3b-clawbench.

sourceHugging Faceapache-2.0updated 4mo agoView on Hugging Face

2likes210downloads

Dataset Card

Qwen3.6-35B-A3B ClawBench Full Results

This repository contains organized ClawBench run artifacts from bench/ClawBench/test-output/qwen36-35b-a3b-full-20260510-2140.

The formal result directories are preserved under results/. Quarantined infrastructure-failure directories and top-level runtime control files such as .pid, .proc, .logpath, and .monitor-hermes-state/ are intentionally excluded from this organized export.

—results/<group>/Qwen3.6-35B-A3B/<run>/: run artifacts including run-meta.json, data/actions.jsonl, data/requests.jsonl, data/agent-messages.jsonl, data/interception.json, screenshots, and recording.mp4 when present.
—results/<group>/batch-logs/: original batch logs for that group.
—metadata/runs.csv: one row per run with parsed run-meta.json fields and relative artifact paths.
—metadata/runs.jsonl: one JSON object per parsed run, including the full run-meta.json payload.
—metadata/aggregate.json: aggregate counts, pass rates, duplicate-case notes, and byte/file totals.
—metadata/missing-run-meta.csv: run directories found without run-meta.json.
—metadata/file_manifest.csv: relative paths and file sizes for the organized export.
—metadata/batch-summaries/: original batch-summary.json files copied into one place.
—source-logs/: top-level experiment and restart logs from the source directory.

Summary

Group	Suite	Run dirs	Parsed runs	Missing metadata	Pass	Intercepted	Pass rate	Bytes
openclaw-v1	v1	159	154	5	20	20	12.99%	7528939435
openclaw-v2	v2	132	130	2	54	54	41.54%	8483743682
hermes-v1	v1	155	153	2	18	18	11.76%	6038334965
hermes-v2	v2	130	130	0	45	45	34.62%	5080753019
Total	-	576	567	9	137	137	24.16%	27131771101

Notes:

—Pass/intercept counts are computed from parsed run-meta.json files, not from batch-summary.json, because some batches were resumed and mark previously completed jobs as skipped in the batch summary.
—openclaw-v1 contains one duplicate parsed case (736-home-services-maintenance-plumbing-ace-hardware) and five run directories without run-meta.json; see metadata/aggregate.json and metadata/missing-run-meta.csv.
—The run artifacts can include synthetic user profile traces and disposable email addresses generated by ClawBench.