File size: 3,455 Bytes
0a0a615 90e48c7 0a0a615 f9c09ba 0a0a615 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 | ---
pretty_name: Fara 1.5 9B 40-Task Baseline Evaluation Artifacts
configs:
- config_name: task-results
data_files: data/task-results-*.parquet
default: true
- config_name: summary
data_files: data/summary-*.parquet
---
# Fara 1.5 9B 40-task baseline evaluation artifacts
This public dataset preserves the complete evaluation artifacts for the 40-task Fara baseline evaluation in WootzApp CUA Bench.
## Result
| Metric | Value |
|---|---:|
| Tasks selected | 40 |
| Passed | 8 |
| Verifier failures | 24 |
| Execution errors | 8 |
| Infrastructure errors | 0 |
| Pass@1 over all selected tasks | 20.0% |
| Pass rate among verified tasks | 25.0% |
Execution errors remain in the denominator. Raw task logs are included so those eight tasks can be audited separately from verifier-scored failures.
## Exact provenance
| Component | Repository or resource | Exact revision |
|---|---|---|
| Upstream model | [`microsoft/Fara1.5-9B`](https://huggingface.co/microsoft/Fara1.5-9B) | `1a93677cd89d5601bc2ed759791e981f3a520032` |
| Preserved model checkpoint | [`WootzappLab/fara-9b-baseline-eval`](https://huggingface.co/WootzappLab/fara-9b-baseline-eval) | `6b15345b4dd7d567863bc08ec3bc79b44780dd99` |
| Fara inference source | Microsoft Fara source used by the serving setup | `a675d6d61c41c47ae87bacefeab22caad18e3e84` |
| Task dataset | [`WootzappLab/cua-bench-baseline-40`](https://huggingface.co/datasets/WootzappLab/cua-bench-baseline-40) | `c2dd9ed8f23cbc477f49be7b7704e42bf4d46de6`; exact snapshot SHA256 `745f113b9c45f1a8dc3581ebc671435a3eafdfdbfb38bcaa6df1097ec10ab2e3` |
| Recorder, browser harness, environments, verifier | [`tokenbender/w8-cua-bench`](https://github.com/tokenbender/w8-cua-bench/tree/environment-work) | `09976c86454f9a87e2fa6a3a3589d2e4d24f180e` |
| Infrastructure code | `infra-changes` branch | `eb5dc3d9e0cbfbe389334b0600002f8db3a795dd` |
| Results bundle | Raw artifact and task snapshot manifest | `43f4ed88b79d9c0abe7a733ac6499b0000364e3f1f676a29320e83d1682e081b` |
The evaluation used GCP instance `w8-fara-9b-8115f370-head-3ibtw2ah-compute` in project `lifeandhalf-24122025`, zone `us-central1-a`.
## Repository layout
- `artifacts/43f4ed88b79d9c0abe7a733ac6499b0000364e3f1f676a29320e83d1682e081b/raw/`: complete raw output, including per-task recordings, model output, logs, rubrics, and verifier results
- `artifacts/43f4ed88b79d9c0abe7a733ac6499b0000364e3f1f676a29320e83d1682e081b/source/tasks.jsonl`: exact 40-task input snapshot
- `data/task-results-00000-of-00001.parquet`: one normalized row per task and the default dataset view
- `data/summary-00000-of-00001.parquet`: evaluation-level summary and provenance
- `metadata/evaluation.json`: machine-readable evaluation receipt and task results
- `metadata/publication.json`: publication identity
- `manifests/43f4ed88b79d9c0abe7a733ac6499b0000364e3f1f676a29320e83d1682e081b.sha256`: SHA256 for every raw artifact and source snapshot
## Integrity check
```bash
sha256sum -c manifests/43f4ed88b79d9c0abe7a733ac6499b0000364e3f1f676a29320e83d1682e081b.sha256
```
The results bundle SHA256 is an integrity checksum derived from a normalized manifest of the raw evaluation output and exact task snapshot.
## Scope
This repository contains evaluation evidence and structured results. It does not contain access tokens, TLS certificates, browser caches, or model weights. The weights are preserved separately in the checkpoint repository linked above.
|