| --- |
| license: cc-by-4.0 |
| language: |
| - en |
| pretty_name: >- |
| quant_eval Golden oracle fixtures |
| size_categories: |
| - 1K<n<10K |
| tags: |
| - quantization |
| - large-language-models |
| - gguf |
| - agent-evaluation |
| - tool-calling |
| - behavioral-evaluation |
| - mcnemar |
| - llama-cpp |
| - model-evaluation |
| - reproducibility |
| annotations_creators: |
| - machine-generated |
| source_datasets: |
| - original |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: fixture_version_crosswalk.csv |
| --- |
| |
| # quant_eval — Golden oracle fixtures |
| |
| **The locked evaluation fixture set — 1,600 cases across eight agent task families with their deterministic ground truth — plus the crosswalk mapping every run's recorded fixture hash and version label to the published file.** |
| |
| Part of the quant_eval public corpus: a per-case behavioral evaluation of full-weight and quantized large language models across eight agent-relevant task families, with paired statistical testing. |
|
|
| **Cite this dataset:** [10.5281/zenodo.22010278](https://doi.org/10.5281/zenodo.22010278) — concept DOI, always resolves to the latest version. |
| **This exact deposit:** [10.5281/zenodo.22010279](https://doi.org/10.5281/zenodo.22010279) — version DOI, frozen. Cite this one where reported numbers must stay verifiable against the object referenced. |
|
|
| ## What this file contains |
|
|
| | File | Rows | Columns | |
| |---|---:|---:| |
| | `golden_oracle_fixtures.json` | 1,600 | — | |
| | `fixture_version_crosswalk.csv` | 6 | 8 | |
|
|
| Supporting files: `source_bundle_checksums.json`. |
|
|
| ## Corpus scope |
|
|
| | Run | Model | Baseline | Quantized | Substrate | Licence | |
| |---|---|---|---|---|---| |
| | `Mistral_Nemo_Instruct_2407_20260814_030505` | mistralai/Mistral-Nemo-Instruct-2407 | gguf_f16 | gguf_q4_k_m | local | Apache-2.0 | |
| | `Mistral_Nemo_Instruct_2407_20260815_113254` | mistralai/Mistral-Nemo-Instruct-2407 | gguf_f16 | gguf_q5_k_m | local | Apache-2.0 | |
| | `Mistral_Nemo_Instruct_2407_20260816_084553` | mistralai/Mistral-Nemo-Instruct-2407 | gguf_f16 | gguf_q8_0 | local | Apache-2.0 | |
| | `Qwen2.5_14B_Instruct_1M_20260815_220633` | Qwen/Qwen2.5-14B-Instruct-1M | modal_f16 | modal_q4_k_m | Modal | Apache-2.0 | |
| | `Qwen2.5_32B_Instruct_20260815_081051` | Qwen/Qwen2.5-32B-Instruct | modal_f16 | modal_q4_k_m | Modal | Apache-2.0 | |
| | `Qwen2.5_7B_Instruct_20260814_234822` | Qwen/Qwen2.5-7B-Instruct | gguf_f16 | gguf_q4_k_m | local | Apache-2.0 | |
|
|
| Every run evaluates a full-weight baseline and a quantized variant of the same model against the identical locked fixture set, case for case. Statistical comparison is paired: the two-sided exact McNemar test on per-case outcomes, with Wilson intervals on the rates. |
|
|
| ## Columns |
|
|
| ### `fixture_version_crosswalk.csv` |
|
|
| ``` |
| run_id fixtures_sha256 fixture_version_label |
| fixture_schema_version evaluation_contract_id fixture_split |
| canonical_content_sha256 is_published_file |
| ``` |
|
|
| ## Verification |
|
|
| This corpus is derived from sanitized publication bundles produced by the quant_eval harness. It is designed to be checked rather than trusted: |
| |
| - `source_bundle_checksums.json`, included here, republishes, verbatim, the SHA-256 digest and byte length of every file in every source bundle. No source file was modified. |
| - Before this file was written, the builder verified all 72 source-file digests and independently recomputed all 96 family x runner pass rates from the raw per-case rows, matching the harness rollups exactly. |
| - Aggregates published elsewhere in this corpus — the paired degradation statistics (D5) and family pass rates (D6) datasets — are recomputable from the per-case results dataset (D1) using the gate definitions in `data_dictionary.json`, published with the per-case results dataset (D1) and the throughput telemetry dataset (D2). |
|
|
| ### Fixture identity |
|
|
| The published fixture file is a **byte-exact copy** of the file used in run `Mistral_Nemo_Instruct_2407_20260816_084553`. |
|
|
| - File SHA-256: `0673b5e5c0fb1b7ba4bdc8c9aa985104260db183453c5708aff14c87823edb97` |
| - Canonical content SHA-256: `b7dd9258ca64c434eb57d32199e610c558a9e24cd0402e419de7ae72624e0cb8` |
|
|
| Across the published runs the fixture file appears under more than one SHA-256 and more than one version label. A full structural comparison of all cases shows the only differing key is the top-level `version` string; all cases, all oracle expectations, and all trace hashes are identical. The canonical content hash is taken over the fixture object with `version` removed, serialized with sorted keys and compact separators, so the equivalence can be reproduced rather than taken on trust. `fixture_version_crosswalk.csv` maps every run to its recorded hash and label, and flags which file is published here. |
|
|
| ## Limits you should know before using this |
|
|
| - **Decoding conditions are not uniform across models.** Temperature follows each publisher's own model card, so cross-model comparison of absolute pass rates is confounded. Within-run pairing is unaffected, which is what the paired test requires. The conditions are published per row and per run so they can be filtered on. |
| - **Runs on the Modal substrate record `seed` status `unsupported`** — the deployed method signature accepts no seed — so those runs are not exactly reproducible. Local runs applied a fixed seed. |
| - **Runtime figures are observed harness wall time** on the recorded hardware and backends. They are not a controlled throughput benchmark and not a general claim about quantization performance at any precision on any hardware. Direction is published as an explicit label because not every measured pair is a speedup. |
| - **The fuzz family is an adaptive trajectory** evaluated from identical starting fixtures. Its paired test compares complete case outcomes, not identical post-divergence prompts. |
| - **Calibration runs are not published.** Runs that informed a published run are disclosed by identifier in `calibration_lineage.csv`, published in the run provenance dataset (D4), so the record is complete without releasing provisional numbers. |
|
|
| ## Citation |
|
|
| ```bibtex |
| @dataset{pbh_quant_eval_d3, |
| author = {Hill, Patrick}, |
| title = {quant_eval Golden oracle fixtures}, |
| publisher = {PBH Applied Systems, LLC}, |
| year = {2026}, |
| doi = {10.5281/zenodo.22010278}, |
| note = {Version DOI: 10.5281/zenodo.22010279}, |
| license = {CC-BY-4.0} |
| } |
| ``` |
|
|
| ## Licence |
|
|
| Creative Commons Attribution 4.0 International (CC BY 4.0). See `LICENSE`. Commercial use is permitted; attribution is required. |
|
|
| This corpus describes third-party models and redistributes no model weights. Each evaluated model remains under its own licence, recorded per run in the run provenance dataset. |
|
|
| --- |
|
|
| Produced by build_datasets.py 2.4.0 from quant_eval publication bundles. Built 2026-08-19. |
|
|
