Hugging Face's logo

Datasets:
pbhappliedsystems
/
quant_eval_efficiency_and_footprint

Modalities:
Tabular
Text
Formats:
csv
Languages:
English
Size:
< 1K
Tags:
quantization
large-language-models
gguf
agent-evaluation
tool-calling
behavioral-evaluation
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
quant_eval_efficiency_and_footprint / README.md
pbhappliedsystems's picture
pbhappliedsystems
Upload 7 files
14f19a7 verified
preview code
|
Raw
History Blame Contribute Delete
7.32 kB
metadata 
license: cc-by-4.0
language:
  - en
pretty_name: quant_eval Efficiency and footprint
size_categories:
  - n<1K
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: quant_eval_efficiency_and_footprint.csv

quant_eval — Efficiency and footprint

One row per published run: stored weight artifact bytes before and after quantization, compression ratio, observed evaluation wall-time ratio with an explicit direction label, and token throughput.

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.22010723 — concept DOI, always resolves to the latest version. This exact deposit: 10.5281/zenodo.22010724 — version DOI, frozen. Cite this one where reported numbers must stay verifiable against the object referenced.

What this file contains

File Rows Columns
quant_eval_efficiency_and_footprint.csv 6 22

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.

Figures

Horizontal bar chart of observed evaluation wall-time ratios for six model-precision pairs. Four local llama.cpp pairs exceed parity at 2.704, 2.564, 1.985, and 1.714 times; two Modal pairs fall below parity at 0.961 and 0.854 times, shown hatched.

Observed evaluation wall-time ratio (full weight divided by quantized) for all six published pairs. Hatched bars fall below parity — the quantized run was slower. Observed harness wall time on the recorded hardware and backends; not a controlled throughput benchmark and not a general claim about quantization performance at any precision on any hardware.

Figures are generated directly from the harness rollups by the published build tooling; no plotted value is recomputed, smoothed, or fitted.

Columns

quant_eval_efficiency_and_footprint.csv 

  run_id                                        model_id                                      baseline_kind
  quant_type                                    baseline_bytes                                quantized_bytes
  compression_ratio                             size_reduction_fraction                       measurement_scope
  excludes                                      baseline_runner                               quantized_runner
  execution_substrate                           baseline_tokens_per_second                    quantized_tokens_per_second
  observed_wall_time_ratio                      wall_time_direction                           throughput_ratio
  token_volume_cost_proxy_ratio                 monetary_cost_ratio_status                    artifact_footprint_status
  hardware_backend_scope

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.
  • The figures here are not derived from the per-case results dataset. Stored-artifact byte counts and observed wall time are recorded by the harness at run time and are carried through from the source bundles unchanged; they are traceable through the bundle digests above, not recomputable from per-case rows. Pass-rate aggregates, which are recomputable, live in the paired degradation statistics (D5) and family pass rates (D6) datasets.

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 

@dataset{pbh_quant_eval_d7,
  author    = {Hill, Patrick},
  title     = {quant_eval Efficiency and footprint},
  publisher = {PBH Applied Systems, LLC},
  year      = {2026},
  doi       = {10.5281/zenodo.22010723},
  note      = {Version DOI: 10.5281/zenodo.22010724},
  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.