Metadata-Version: 2.4
Name: evalvitals
Version: 0.1.0
Summary: Failure case analysis in the era of LLM.
License: Creative Commons Legal Code
        
        CC0 1.0 Universal
        
            CREATIVE COMMONS CORPORATION IS NOT A LAW FIRM AND DOES NOT PROVIDE
            LEGAL SERVICES. DISTRIBUTION OF THIS DOCUMENT DOES NOT CREATE AN
            ATTORNEY-CLIENT RELATIONSHIP. CREATIVE COMMONS PROVIDES THIS
            INFORMATION ON AN "AS-IS" BASIS. CREATIVE COMMONS MAKES NO WARRANTIES
            REGARDING THE USE OF THIS DOCUMENT OR THE INFORMATION OR WORKS
            PROVIDED HEREUNDER, AND DISCLAIMS LIABILITY FOR DAMAGES RESULTING FROM
            THE USE OF THIS DOCUMENT OR THE INFORMATION OR WORKS PROVIDED
            HEREUNDER.
        
        Statement of Purpose
        
        The laws of most jurisdictions throughout the world automatically confer
        exclusive Copyright and Related Rights (defined below) upon the creator
        and subsequent owner(s) (each and all, an "owner") of an original work of
        authorship and/or a database (each, a "Work").
        
        Certain owners wish to permanently relinquish those rights to a Work for
        the purpose of contributing to a commons of creative, cultural and
        scientific works ("Commons") that the public can reliably and without fear
        of later claims of infringement build upon, modify, incorporate in other
        works, reuse and redistribute as freely as possible in any form whatsoever
        and for any purposes, including without limitation commercial purposes.
        These owners may contribute to the Commons to promote the ideal of a free
        culture and the further production of creative, cultural and scientific
        works, or to gain reputation or greater distribution for their Work in
        part through the use and efforts of others.
        
        For these and/or other purposes and motivations, and without any
        expectation of additional consideration or compensation, the person
        associating CC0 with a Work (the "Affirmer"), to the extent that he or she
        is an owner of Copyright and Related Rights in the Work, voluntarily
        elects to apply CC0 to the Work and publicly distribute the Work under its
        terms, with knowledge of his or her Copyright and Related Rights in the
        Work and the meaning and intended legal effect of CC0 on those rights.
        
        1. Copyright and Related Rights. A Work made available under CC0 may be
        protected by copyright and related or neighboring rights ("Copyright and
        Related Rights"). Copyright and Related Rights include, but are not
        limited to, the following:
        
          i. the right to reproduce, adapt, distribute, perform, display,
             communicate, and translate a Work;
         ii. moral rights retained by the original author(s) and/or performer(s);
        iii. publicity and privacy rights pertaining to a person's image or
             likeness depicted in a Work;
         iv. rights protecting against unfair competition in regards to a Work,
             subject to the limitations in paragraph 4(a), below;
          v. rights protecting the extraction, dissemination, use and reuse of data
             in a Work;
         vi. database rights (such as those arising under Directive 96/9/EC of the
             European Parliament and of the Council of 11 March 1996 on the legal
             protection of databases, and under any national implementation
             thereof, including any amended or successor version of such
             directive); and
        vii. other similar, equivalent or corresponding rights throughout the
             world based on applicable law or treaty, and any national
             implementations thereof.
        
        2. Waiver. To the greatest extent permitted by, but not in contravention
        of, applicable law, Affirmer hereby overtly, fully, permanently,
        irrevocably and unconditionally waives, abandons, and surrenders all of
        Affirmer's Copyright and Related Rights and associated claims and causes
        of action, whether now known or unknown (including existing as well as
        future claims and causes of action), in the Work (i) in all territories
        worldwide, (ii) for the maximum duration provided by applicable law or
        treaty (including future time extensions), (iii) in any current or future
        medium and for any number of copies, and (iv) for any purpose whatsoever,
        including without limitation commercial, advertising or promotional
        purposes (the "Waiver"). Affirmer makes the Waiver for the benefit of each
        member of the public at large and to the detriment of Affirmer's heirs and
        successors, fully intending that such Waiver shall not be subject to
        revocation, rescission, cancellation, termination, or any other legal or
        equitable action to disrupt the quiet enjoyment of the Work by the public
        as contemplated by Affirmer's express Statement of Purpose.
        
        3. Public License Fallback. Should any part of the Waiver for any reason
        be judged legally invalid or ineffective under applicable law, then the
        Waiver shall be preserved to the maximum extent permitted taking into
        account Affirmer's express Statement of Purpose. In addition, to the
        extent the Waiver is so judged Affirmer hereby grants to each affected
        person a royalty-free, non transferable, non sublicensable, non exclusive,
        irrevocable and unconditional license to exercise Affirmer's Copyright and
        Related Rights in the Work (i) in all territories worldwide, (ii) for the
        maximum duration provided by applicable law or treaty (including future
        time extensions), (iii) in any current or future medium and for any number
        of copies, and (iv) for any purpose whatsoever, including without
        limitation commercial, advertising or promotional purposes (the
        "License"). The License shall be deemed effective as of the date CC0 was
        applied by Affirmer to the Work. Should any part of the License for any
        reason be judged legally invalid or ineffective under applicable law, such
        partial invalidity or ineffectiveness shall not invalidate the remainder
        of the License, and in such case Affirmer hereby affirms that he or she
        will not (i) exercise any of his or her remaining Copyright and Related
        Rights in the Work or (ii) assert any associated claims and causes of
        action with respect to the Work, in either case contrary to Affirmer's
        express Statement of Purpose.
        
        4. Limitations and Disclaimers.
        
         a. No trademark or patent rights held by Affirmer are waived, abandoned,
            surrendered, licensed or otherwise affected by this document.
         b. Affirmer offers the Work as-is and makes no representations or
            warranties of any kind concerning the Work, express, implied,
            statutory or otherwise, including without limitation warranties of
            title, merchantability, fitness for a particular purpose, non
            infringement, or the absence of latent or other defects, accuracy, or
            the present or absence of errors, whether or not discoverable, all to
            the greatest extent permissible under applicable law.
         c. Affirmer disclaims responsibility for clearing rights of other persons
            that may apply to the Work or any use thereof, including without
            limitation any person's Copyright and Related Rights in the Work.
            Further, Affirmer disclaims responsibility for obtaining any necessary
            consents, permissions or other rights required for any use of the
            Work.
         d. Affirmer understands and acknowledges that Creative Commons is not a
            party to this document and has no duty or obligation with respect to
            this CC0 or use of the Work.
        
Project-URL: Homepage, https://github.com/evalvitals/evalvitals
Project-URL: Documentation, https://evalvitals.github.io/evalvitals/
Project-URL: Repository, https://github.com/evalvitals/evalvitals
Project-URL: Issues, https://github.com/evalvitals/evalvitals/issues
Project-URL: Changelog, https://github.com/evalvitals/evalvitals/blob/main/CHANGELOG.md
Keywords: llm,vlm,interpretability,failure-analysis,explainability
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Science/Research
Classifier: License :: CC0 1.0 Universal (CC0 1.0) Public Domain Dedication
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: pyyaml>=6.0
Requires-Dist: numpy>=1.24
Provides-Extra: local
Requires-Dist: torch>=2.0.0; extra == "local"
Requires-Dist: transformers<5.0.0,>=4.51.0; extra == "local"
Requires-Dist: accelerate>=0.26.0; extra == "local"
Provides-Extra: api
Requires-Dist: openai>=1.0; extra == "api"
Provides-Extra: interp
Requires-Dist: captum>=0.7; extra == "interp"
Requires-Dist: inseq>=0.6; extra == "interp"
Requires-Dist: nnsight>=0.3; extra == "interp"
Provides-Extra: data
Requires-Dist: datasets>=2.0; extra == "data"
Requires-Dist: pillow>=9.0; extra == "data"
Provides-Extra: viz
Requires-Dist: matplotlib>=3.7; extra == "viz"
Provides-Extra: stats
Requires-Dist: statsmodels>=0.14; extra == "stats"
Requires-Dist: scikit-learn>=1.3; extra == "stats"
Provides-Extra: dashboard
Requires-Dist: streamlit>=1.30; extra == "dashboard"
Requires-Dist: pandas>=2.0; extra == "dashboard"
Provides-Extra: gemini
Requires-Dist: google-genai>=1.0; extra == "gemini"
Provides-Extra: all
Requires-Dist: torch>=2.0.0; extra == "all"
Requires-Dist: transformers<5.0.0,>=4.51.0; extra == "all"
Requires-Dist: accelerate>=0.26.0; extra == "all"
Requires-Dist: openai>=1.0; extra == "all"
Requires-Dist: captum>=0.7; extra == "all"
Requires-Dist: inseq>=0.6; extra == "all"
Requires-Dist: nnsight>=0.3; extra == "all"
Requires-Dist: matplotlib>=3.7; extra == "all"
Requires-Dist: statsmodels>=0.14; extra == "all"
Requires-Dist: scikit-learn>=1.3; extra == "all"
Requires-Dist: streamlit>=1.30; extra == "all"
Requires-Dist: pandas>=2.0; extra == "all"
Provides-Extra: dev
Requires-Dist: pytest>=8.0; extra == "dev"
Requires-Dist: pytest-cov>=4.0; extra == "dev"
Requires-Dist: ruff>=0.4; extra == "dev"
Requires-Dist: mypy>=1.8; extra == "dev"
Requires-Dist: jsonschema>=4.0; extra == "dev"
Dynamic: license-file

# EvalVitals

[![CI](https://github.com/evalvitals/evalvitals/actions/workflows/ci.yml/badge.svg)](https://github.com/evalvitals/evalvitals/actions/workflows/ci.yml)
[![PyPI](https://img.shields.io/pypi/v/evalvitals)](https://pypi.org/project/evalvitals/)
[![Python](https://img.shields.io/pypi/pyversions/evalvitals)](https://pypi.org/project/evalvitals/)
[![Docs](https://img.shields.io/badge/docs-mkdocs-blue)](https://evalvitals.github.io/evalvitals/)
[![License: CC0-1.0](https://img.shields.io/badge/license-CC0--1.0-green)](LICENSE)

EvalVitals is an sklearn-like toolkit for failure-case analysis in the era of
LLMs, VLMs, omni (text+image+audio+video) models, and agentic systems. Model
identity (`ModelSpec`) is separate from runtime (`Backend`); analyzers
(`Analyzer`) connect to models by capability matching (`Capability`), so the
same spec runs through a black-box API or a white-box local backend — only
the capability set changes.

| Contract | Role |
|---|---|
| `ModelSpec` | Model identity: family, HF repo, architecture traits, modalities. |
| `Backend` | Runtime: local HF internals, black-box API, or offline batch engines. |
| `Model` | A runnable object with `generate()` and `forward(capture=...) -> Trace`. |
| `Analyzer` | An sklearn-style estimator: `Analyzer(**params).run(model, data) -> Result`. |
| `Capability` | Vocabulary matching analyzers to compatible model runtimes. |
| `FailureCase` | Central data unit for prompts, labels, provenance, agent trajectories. |
| `Result` | Uniform output: human-readable summary + structured findings. |

## What's Inside

EvalVitals covers three layers of the analysis workflow, usable independently
or chained into one automated loop:

1. **Analyzer toolkit** — 26 registered analyzers (attention, uncertainty,
   hallucination, Shapley attribution, logit lens, representation geometry,
   agent-trajectory analysis) run against any model/backend pair through the
   same `Analyzer(**params).run(model, data) -> Result` call shape. See the
   [Demo](#demo) below and the [Analyzer Zoo](docs/analyzers.md).
2. **Data analysis agent (M2/M3)** — `evalvitals explore` points a coding
   agent at a raw results directory (any JSON/JSONL shape, no host-side
   parsing) and gets back descriptive takeaways, charts, and 1-3 falsifiable
   hypotheses — no code required. See
   [Exploratory Analysis (M2/M3)](docs/m2_analysis.md).
3. **Intervention (M4/M5)** — `HypothesisTester` verifies a hypothesis
   statistically and against the stated experiment protocol; `FixAgent` then
   proposes and validates candidate repairs (prompt → scaffold → internals →
   parameter space) against the unmodified baseline with paired McNemar +
   e-value. See [Intervention & Verification (M4/M5)](docs/intervention.md).

`VLDiagnoseLoop` chains all three (M1→M2→M3→M5, M4 post-loop) into one
automated failure-attribution pipeline — see [Quickstart](docs/quickstart.md).

## Demo

Relative attention on a VLM — ["MLLMs Know Where to Look"](https://arxiv.org/abs/2502.17422) ([code](https://github.com/saccharomycetes/mllms_know)):

```python
from PIL import Image
from evalvitals import compose, Capability
from evalvitals.analyzers.attention import RelativeAttentionAnalyzer
from evalvitals.core.case import Inputs

# Load Qwen2.5-VL with white-box attention capture
model = compose("qwen2.5-vl-7b-instruct", "hf_local", want={Capability.ATTENTION})

# Run relative attention: ratio of task-specific vs generic image attention
result = RelativeAttentionAnalyzer(layer=22, top_k=5).run(
    model,
    Inputs(prompt="What color is the car?", image=Image.open("scene.jpg")),
)

print(result.summary())   # agent-readable findings
result.plot()             # (H, W) heatmap — requires evalvitals[viz]
```

The same call shape works for any registered model/analyzer pair — a plain
text LLM, a config-driven YAML run, or an explicit backend — see
[Quickstart](docs/quickstart.md) for those and for the automated
failure-attribution loop (`VLDiagnoseLoop`, M1→M2→M3→M5) and no-code
exploratory analysis (`evalvitals explore`).

## Install

```bash
pip install -e .
pip install -e ".[viz]"
pip install -e ".[dev]"
```

## Documentation

- [Docs overview](docs/index.md)
- [Quickstart](docs/quickstart.md) — runnable examples for every entry point, the diagnosis loop, and submitting a run
- [Exploratory Analysis (M2/M3)](docs/m2_analysis.md) — standalone `evalvitals explore`
- [Intervention & Verification (M4/M5)](docs/intervention.md) — `HypothesisTester` verification, `FixAgent` tiered repair
- [Analyzer Zoo](docs/analyzers.md) — reference tables of analyzers and registered models
- [Architecture](docs/architecture.md) — package structure and design contracts
- [Extending EvalVitals](docs/extending.md) — add analyzers, specs, backends
- [Roadmap](docs/roadmap.md)

## Examples

Each directory under `examples/` is a self-contained, runnable demo:

```bash
cd examples/analyzer_demos/qwen_attention  && docker compose up   # attention analysis on a text LLM
cd examples/m2_statistics/deco_hallu_explore && docker compose up # M2/M3 explore, real M1 data
cd examples/m2_statistics/deco_hallu_explore && bash run_attn.sh  # ... attention-enriched: FAIL/PASS distributions + cross-checkpoint geometry (no GPU; data ships with the repo)
cd examples/m2_statistics/deco_hallu_explore && bash run_attn_pipeline.sh  # ... FULL held-out pipeline: propose → held-out test + LLM judge → L1..L3b fix → one 4-tab report (SKIP_FIX=1 for the no-GPU half)
cd examples/diagnosis_loops/qwen_loop_agy  && docker compose up   # VLDiagnoseLoop M1→M5 (VLM)
```

See [`examples/README.md`](examples/README.md) for the full list, grouped by
layer (single analyzers, standalone M2/M3, full diagnosis loop).
