diffeval 0.1.0__py3-none-any.whl

diffeval-0.1.0.dist-info/METADATA

@@ -0,0 +1,378 @@
+Metadata-Version: 2.4
+Name: diffeval
+Version: 0.1.0
+Summary: Evaluation harness for image and video diffusion models
+Author: woletee
+License-Expression: Apache-2.0
+Keywords: diffusion,evaluation,benchmark,text-to-image,fid,kid,cmmd,clip-score
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Scientific/Engineering :: Image Recognition
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: pyyaml
+Requires-Dist: pillow
+Requires-Dist: numpy
+Provides-Extra: dev
+Requires-Dist: pytest>=8; extra == "dev"
+Provides-Extra: diffusers
+Requires-Dist: torch; extra == "diffusers"
+Requires-Dist: diffusers; extra == "diffusers"
+Requires-Dist: transformers; extra == "diffusers"
+Requires-Dist: accelerate; extra == "diffusers"
+Requires-Dist: safetensors; extra == "diffusers"
+Provides-Extra: metrics
+Requires-Dist: torch; extra == "metrics"
+Requires-Dist: transformers; extra == "metrics"
+Requires-Dist: clean-fid; extra == "metrics"
+Requires-Dist: scipy<1.18; extra == "metrics"
+Provides-Extra: benchmarks
+Requires-Dist: pyarrow; extra == "benchmarks"
+Requires-Dist: huggingface_hub; extra == "benchmarks"
+Provides-Extra: geneval
+Provides-Extra: video
+Dynamic: license-file
+
+<div align="center">
+
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/woletee/diffeval/main/docs/assets/logo-dark.svg">
+  <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/woletee/diffeval/main/docs/assets/logo-light.svg">
+  <img alt="diffeval" src="https://raw.githubusercontent.com/woletee/diffeval/main/docs/assets/logo-light.svg" width="240">
+</picture>
+
+# diffeval: A Framework for Diffusion Model Evaluation
+
+A fixed prompt and seed suite, scored the same way on every model.
+
+[![CI](https://img.shields.io/github/actions/workflow/status/woletee/diffeval/ci.yml?style=flat&label=CI&colorA=000000&colorB=000000)](https://github.com/woletee/diffeval/actions/workflows/ci.yml)
+[![Python](https://img.shields.io/badge/python-3.10%2B-000000?style=flat&colorA=000000&colorB=000000)](https://github.com/woletee/diffeval/blob/main/pyproject.toml)
+[![License](https://img.shields.io/badge/license-Apache--2.0-000000?style=flat&colorA=000000&colorB=000000)](https://github.com/woletee/diffeval/blob/main/LICENSE)
+
+<p>
+  <a href="#quickstart">Quickstart</a>
+  ·
+  <a href="#metrics">Metrics</a>
+  ·
+  <a href="#backends">Backends</a>
+  ·
+  <a href="#suites">Suites</a>
+  ·
+  <a href="#extending-diffeval">Extending</a>
+  ·
+  <a href="docs/DESIGN.md">Design</a>
+</p>
+
+</div>
+
+diffeval is an evaluation harness for image and video diffusion models, in the same role [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) has for language models. You name a backend and a suite. The suite fixes the prompts, the seeds, and the resolution. For each seed the backend writes one image per prompt, every metric scores that seed, and the runner aggregates a mean and a sample standard deviation. The run writes a console table, `results.json`, `report.md`, and `report.html`.
+
+Video scoring is later. This version scores images. The `video` extra installs no packages, and there is no `vbench` metric.
+
+- **Pluggable.** Suites, metrics, and backends register by name. `diffeval list` and `diffeval run` look names up in `TASKS`, `METRICS`, and `BACKENDS`.
+- **Light by default.** The base install is PyYAML, Pillow, and NumPy. Torch, Diffusers, and clean-fid stay in optional extras.
+- **Honest scores.** A missing library or a missing reference is status `skipped`, with the reason stored on the row. The images stay on disk.
+- **Traceable.** `results.json` uses `schema_version` 1. It stores the suite, `describe()`, package versions, the GPU when torch reports one, and `metric_protocols`.
+
+> [!NOTE]
+> diffeval publishes no benchmark numbers. The bundled `core` suite is a smoke check. The `diffusers` backend has not been validated on a GPU in this repository. Compare a score with a paper only after matching that paper's reference implementation and dataset.
+
+## Installation
+
+diffeval requires Python 3.10 or newer.
+
+```bash
+pip install diffeval
+```
+
+Add extras in brackets, for example `pip install "diffeval[diffusers,metrics]"`. To work on the code, clone the repository and run `pip install -e ".[dev]"`.
+
+That installs the CLI, the YAML loader, Pillow, and NumPy.
+
+| Extra | Installs | Used for |
+| --- | --- | --- |
+| `diffusers` | torch, diffusers, transformers, accelerate, safetensors | the `diffusers` backend |
+| `metrics` | torch, transformers, clean-fid, scipy<1.18 | `clip_score`, `cmmd`, `fid`, `kid` |
+| `benchmarks` | pyarrow, huggingface_hub | `diffeval-coco30k-images` |
+| `dev` | pytest | the tests |
+| `geneval` | nothing | keeps the GenEval detector off the default install |
+| `video` | nothing | keeps VBench off the default install |
+
+```bash
+pip install -e ".[diffusers,metrics]"
+pip install -e ".[dev]"
+```
+
+The `geneval` metric returns `skipped`. The `video` extra does not install VBench.
+
+## Quickstart
+
+Start with a CPU smoke run. The `dummy` backend draws deterministic RGB images, so the command downloads nothing.
+
+```bash
+diffeval run --backend dummy --suite core
+```
+
+If you omit both `--backend` and `--pipeline`, the backend is `dummy`. `--pipeline` without `--backend` selects `diffusers`.
+
+Then a Diffusers model. Install the `diffusers` extra first. The command downloads weights. It has not been run in this repository, and CI does not run it. Without the extra, the process prints `install the diffusers extra: pip install 'diffeval[diffusers]'` and exits 2.
+
+One smaller model has been run on a CPU. `stabilityai/sd-turbo` with `--steps 1 --guidance 0 --device cpu` generated all eight `core` images and scored `clip_score`, `cmmd`, `fid`, and `kid`, at about five seconds per image. That run says the backend and metrics work end to end. Its scores against the four `core` drawings are not quality numbers.
+
+```bash
+diffeval run --pipeline stabilityai/sd-turbo --suite core --steps 1 --guidance 0 --device cpu
+```
+
+```bash
+diffeval run --pipeline black-forest-labs/FLUX.1-schnell --suite core
+```
+
+When `--steps` and `--guidance` are omitted, `black-forest-labs/FLUX.1-schnell` uses 4 steps and guidance 0. Any other model id leaves the pipeline defaults in place. A value passed on the CLI wins.
+
+Each run writes a directory. The default is `outputs/<UTC timestamp>-<suite>-<backend>`. The timestamp looks like `20261025T153000Z`. Spaces and slashes in that label become underscores. The model id stays in `describe()`. The folder name ends with the backend, `dummy` or `diffusers`.
+
+```text
+outputs/20261025T153000Z-core-dummy/
+├── images/
+│   ├── seed_0/red_cube.png
+│   └── seed_1/red_cube.png
+├── results.json
+├── report.md
+└── report.html
+```
+
+`core` has four prompts, so each seed directory holds four PNGs. Paths inside `results.json` are relative to the output directory, for example `images/seed_0/red_cube.png`.
+
+The console table and `report.md` have one row per metric. Columns are metric, status, mean, std, per-seed, and direction. With only the base install, the metrics are `skipped`:
+
+```text
+metric      status   mean  std  per-seed  direction
+clip_score  skipped  n/a   n/a  n/a,n/a   higher
+cmmd        skipped  n/a   n/a  n/a,n/a   lower
+fid         skipped  n/a   n/a  n/a,n/a   lower
+kid         skipped  n/a   n/a  n/a,n/a   lower
+```
+
+Mean is the arithmetic mean of seeds whose status is `ok`. Standard deviation is the sample standard deviation, `ddof=1`. It is null when fewer than two seeds succeeded, and the console prints that null as `n/a`. The row stores `std_ddof` as 1. `core` uses two seeds so a real metric can show a standard deviation.
+
+`report.html` repeats the table and shows an image grid grouped by prompt, one figure per seed. Prompts are escaped. The page says to check scores against the published reference implementation and dataset, and it calls `core` a smoke suite.
+
+Exit code 0 means the report was written. Exit code 2 means the suite, backend, or metric name was unknown, generation failed, the chosen backend's extra is missing, or `--strict` found a metric that is not `ok`. `--strict` is off by default. The runner writes the report first, then exits 2. A smoke run with the metrics extra missing still exits 0.
+
+```bash
+diffeval list tasks
+diffeval list metrics
+diffeval list backends
+diffeval run --backend dummy --suite core --seeds 0 --output outputs/smoke
+diffeval run --backend dummy --suite core --metrics geneval --strict
+```
+
+The last command writes the report and exits 2. `geneval` is skipped, and `--strict` requires every aggregate status to be `ok`.
+
+## Command reference
+
+```bash
+diffeval list tasks
+diffeval list metrics
+diffeval list backends
+diffeval run --suite NAME_OR_PATH [options]
+```
+
+| Option | Description |
+| --- | --- |
+| `--suite` | A registered suite name, or a path ending in `.yaml` or `.yml`. Required. A bare name is looked up in `TASKS`. |
+| `--backend` | `dummy` or `diffusers`. Defaults to `diffusers` when `--pipeline` is set, otherwise `dummy`. |
+| `--pipeline` | Hugging Face model id. Required for the `diffusers` backend. |
+| `--steps`, `--guidance` | Sampler steps and guidance scale. FLUX.1-schnell presets to 4 steps and guidance 0. Other models use the pipeline defaults. CLI values win. |
+| `--dtype` | Requested dtype. On CPU, float16 and bfloat16 load as float32. `describe()` stores the dtype actually used in `dtype` and the flag in `requested_dtype`. |
+| `--device` | Device for the pipeline and for metrics. With no flag, metrics use CUDA when torch reports it, otherwise MPS, otherwise CPU. Without torch, metrics use CPU. |
+| `--attention` | `sdpa`, `xformers`, or `eager`. With no flag, the pipeline default stays. |
+| `--variant` | Diffusers pipeline variant. |
+| `--batch-size` | Positive integer. Default 1. |
+| `--seeds` | Comma-separated integers, overriding the suite. Example: `--seeds 0,1,2`. |
+| `--metrics` | Comma-separated metric names, overriding the suite. |
+| `--width`, `--height` | Image size in pixels, overriding the suite. |
+| `--output` | Output directory. |
+| `--strict` | Exit 2 unless every aggregate status is `ok`. The report is still written. |
+
+`describe()` also records `steps_source` and `guidance_source` as `cli`, `preset`, or `pipeline-default`.
+
+## Metrics
+
+| Metric | Better | Implementation | Needs references |
+| --- | :---: | --- | :---: |
+| `clip_score` | higher | Hessel et al. 2021, CLIP ViT-B/32. Mean of `2.5 * max(cosine, 0)` on the raw prompt, truncated to 77 tokens. | no |
+| `cmmd` | lower | Jayasumana et al. 2024, CLIP ViT-L/14 at 336, center square crop, L2-normalized vision embeddings. Minimum-variance RBF MMD, diagonal included, sigma 10, scaled by 1000. Matches the [cmmd-pytorch](https://github.com/sayakpaul/cmmd-pytorch) sample check of 7.696. | yes |
+| `fid` | lower | clean-fid `compute_fid`, `mode="clean"`, Inception-v3. clean-fid resizes each RGB image to 299×299. diffeval does not resize the PNG first. | yes |
+| `kid` | lower | clean-fid `compute_kid` on the same features. Degree-3 polynomial, 100 subsets, subset size at most 1000. | yes |
+| `geneval` | higher | Stub. Returns `skipped`. `details.by_tag` maps `single_object`, `two_object`, `counting`, `colors`, `position`, and `color_attr` to null. Mask2Former is not imported. | no |
+
+The registered metrics are `clip_score`, `cmmd`, `fid`, `kid`, and `geneval`.
+
+`clip_score`, `cmmd`, `fid`, and `kid` return `skipped` when their library is missing. `fid`, `kid`, and `cmmd` also skip when the suite has no reference. A reference is an image folder, precomputed `clean_fid_stats` for `fid` and `kid`, or a `.npy` CLIP embedding for `cmmd`. `fid` and `kid` also skip when the generated folder or the reference folder has fewer than two images. Both calls seed NumPy with the evaluation seed and restore the previous state. FID, KID, and CMMD set `details.warning` when the image count is below the published COCO setup of 30000. The `metrics` extra pins SciPy below 1.18: clean-fid 0.1.35 calls `scipy.linalg.sqrtm` with `disp=False`, which SciPy 1.18 removed, so `fid` and `kid` would error.
+
+`diffeval-check-cmmd` downloads the six cmmd-pytorch sample images and checks that `cmmd` returns 7.696, the value that implementation documents. Earlier versions skipped the L2 normalization and the center crop, so their `cmmd` values are not comparable with published ones.
+
+[Benchmarks](https://github.com/woletee/diffeval/blob/main/docs/BENCHMARKS.md) compares diffeval with published CMMD values and lists the commands to reproduce them.
+
+Each metric's `protocol` dict is copied into `metric_protocols` in `results.json`.
+
+### How scores are aggregated
+
+| Status | Meaning |
+| --- | --- |
+| `ok` | every seed succeeded |
+| `partial` | some seeds succeeded |
+| `skipped` | every seed was skipped |
+| `error` | at least one seed failed and none succeeded |
+
+The mean uses successful seeds only. The standard deviation is null below two successful seeds, which the console prints as `n/a`.
+
+## Backends
+
+| Backend | Description |
+| --- | --- |
+| `dummy` | Deterministic RGB images. CI and the smoke command use it. `describe()` sets `device` to `cpu` and `attention_backend` to `none`. |
+| `diffusers` | `AutoPipelineForText2Image`. Each request gets its own CPU `torch.Generator`, seeded from that request. |
+
+fal and Replicate are not registered. v1 ships `dummy` and `diffusers` only.
+
+## Suites
+
+A suite is a YAML file. A file that omits width or height gets 512 for each. A file that omits metrics gets `clip_score`, `cmmd`, `fid`, and `kid`. Prompt ids match `[A-Za-z0-9_-]+`, because they become file names.
+
+```yaml
+name: shapes
+description: Two prompts for a local check.
+seeds: [0, 1]
+width: 512
+height: 512
+metrics: [clip_score]
+prompts:
+  - {id: red_cube, prompt: a red cube}
+  - {id: blue_sphere, prompt: a blue sphere}
+```
+
+```bash
+diffeval run --backend dummy --suite path/to/shapes.yaml
+```
+
+The bundled `core` suite has four synthetic prompts (`red_cube`, `blue_sphere`, `green_triangle`, `yellow_star`), seeds `[0, 1]`, 512×512 images, and small reference drawings under `reference_dir`. Those drawings are not COCO. A `core` score is a smoke check.
+
+## Extending diffeval
+
+Importing `diffeval` loads the plugins. Each task, metric, and backend registers into `TASKS`, `METRICS`, or `BACKENDS`. Add a module, import it from the matching package `__init__.py`, and the name shows up in `diffeval list`.
+
+<details>
+<summary><b>Add a metric</b></summary>
+
+<br>
+
+The runner constructs the class with no arguments and calls `compute(samples, context)` once per seed. Return a `MetricResult`. Status `ok` carries a finite float. Status `skipped` or `error` carries a null value and an error string. `protocol` is copied into `metric_protocols`.
+
+```python
+from diffeval.metrics.ok import ok
+from diffeval.metrics.skipped import skipped
+from diffeval.registry.register import register
+from diffeval.registry.tables import METRICS
+
+
+@register(METRICS, "example")
+class Example:
+    name = "example"
+    description = "What the number means."
+    higher_is_better = True
+    requires_references = False
+    protocol = {"formula": "state the formula here"}
+
+    def compute(self, samples, context):
+        try:
+            import some_library
+        except ImportError:
+            return skipped(self.name, "install some_library", self.higher_is_better, self.protocol)
+        value = ...  # a finite float for this seed
+        return ok(self.name, value, self.higher_is_better, self.protocol)
+```
+
+A missing optional library returns `skipped`, with the reason in the error string.
+
+</details>
+
+<details>
+<summary><b>Add a backend</b></summary>
+
+<br>
+
+The runner calls `from_config(config)` on the class, then `prepare()`, `generate(requests)`, and `describe()`.
+
+`generate` returns one `diffeval.types.sample.Sample` per request, in request order. `Sample.image` is a PIL RGB image seeded from that request's seed. The runner writes it to a path like `images/seed_0/red_cube.png`.
+
+`describe()` returns `name`, `pipeline`, `dtype`, `requested_dtype`, `steps`, `guidance`, `steps_source`, `guidance_source`, `batch_size`, `device`, `attention_backend`, and `variant`.
+
+[`src/diffeval/backends/dummy/backend.py`](src/diffeval/backends/dummy/backend.py) is the short example.
+
+</details>
+
+<details>
+<summary><b>Register a named suite</b></summary>
+
+<br>
+
+To make a suite available by name, register an object with `description` and `load()`. The runner calls `load()` on that object. It does not construct the class. Follow [`src/diffeval/tasks/core.py`](src/diffeval/tasks/core.py), which registers an instance.
+
+```python
+from diffeval.registry.register import register
+from diffeval.registry.tables import TASKS
+from diffeval.tasks.load_path import load_path
+
+
+class Shapes:
+    description = "Two prompts for a local check."
+
+    def load(self):
+        return load_path("path/to/shapes.yaml")
+
+
+register(TASKS, "shapes")(Shapes())
+```
+
+To ship a suite with the package, put its YAML in `src/diffeval/tasks/suites/` and import the module from `src/diffeval/tasks/__init__.py`.
+
+</details>
+
+## Project layout
+
+```text
+src/diffeval/
+├── cli/        argument parsing and the list / run commands
+├── tasks/      suite loading and the bundled suites
+├── backends/   dummy and diffusers
+├── metrics/    clip_score, cmmd, fid, kid, geneval
+├── runner/     per-seed generation and scoring
+├── aggregate/  mean, sample std, and status across seeds
+├── report/     console, Markdown, and HTML
+├── registry/   TASKS, METRICS, and BACKENDS
+├── system/     package versions, GPU, and attention flags
+└── types/      shared records
+```
+
+Every `.py` file in `src/` and `tests/` has at most 20 lines of code. Blank lines, comments, and docstrings do not count. Imports do. [docs/DESIGN.md](https://github.com/woletee/diffeval/blob/main/docs/DESIGN.md) is the rule and the module tree. [docs/ROADMAP.md](https://github.com/woletee/diffeval/blob/main/docs/ROADMAP.md) is the split into pull requests.
+
+## Contributing
+
+```bash
+pip install -e ".[dev]"
+pytest
+```
+
+CI runs that on Python 3.10, 3.11, and 3.12, then runs `diffeval run --backend dummy --suite core`.
+
+Keep new behavior inside the line cap. Register tasks, metrics, and backends by name. Do not commit model weights. Do not report a smoke score as a paper result.
+
+## License
+
+Apache-2.0. See [LICENSE](https://github.com/woletee/diffeval/blob/main/LICENSE).

diffeval-0.1.0.dist-info/RECORD

@@ -0,0 +1,185 @@
+diffeval/__init__.py,sha256=AV7OXV_zliA85zXZp0JaD58eH3-diIwaipSbrtka8nA,160
+diffeval/__main__.py,sha256=hONcUGf6l2JDkNEOBBs6owvrP91THESdcvb-OKTyXY4,124
+diffeval/plugins.py,sha256=6YwbM0KrBKUL1PdIfzPdo2aajFjo7wioOiXXkAWWJts,164
+diffeval/aggregate/__init__.py,sha256=KwYK6PHc5s55U6gRx4eHlC5TZMfWKtRIwCl87F6AZ3s,56
+diffeval/aggregate/mean.py,sha256=2aRWBo__UbVi8yhZtct7m2h0v-lKiKqMqlF4RNuiNzQ,160
+diffeval/aggregate/names.py,sha256=PEfIVNyhQE6DALeIOq92otbnaGU63QETVWYHpkTx5Lw,307
+diffeval/aggregate/row.py,sha256=gsKzRqkOZNIAap-XUk_FB7OkI2uNd79LZUDzQzR9l5Q,836
+diffeval/aggregate/sample_std.py,sha256=5_xCdWQu0S7kCwsN3WJUYmxvef_AMEkE3tH-ONrrCnU,203
+diffeval/aggregate/series.py,sha256=J8mQeyZ5fSC9FQGy25_3zR62ZNa4TKUXTfAPDJy1kJk,811
+diffeval/aggregate/table.py,sha256=oHwG6hC8LA6JT2QhGPsOWVw1kIOb0sKbo4BKGVA93Yg,255
+diffeval/backends/__init__.py,sha256=W_ADkvPCD4W43oLqzDcuVMv4x7MnzfOyzuTFnMCpfBY,127
+diffeval/backends/noop_prepare.py,sha256=_LS3rx81zdA5sOf3DhZFcJFhUlKMD0ST68Yhenc3vj8,79
+diffeval/backends/protocol.py,sha256=qjBjzXjtMjIPeY8xXHjoRl6G--W2s55pgwnEqRfQDv0,420
+diffeval/backends/diffusers/__init__.py,sha256=a4QTpEH45OA_ca6ki9W3PYsPPWQN3pSnirG_Krbcl2E,40
+diffeval/backends/diffusers/attention_apply.py,sha256=sDsrWtcuOcDv26ReWucuf5lYba3KOAcsCEHtCQWoUmA,596
+diffeval/backends/diffusers/attention_eager.py,sha256=SRPOG1h_9T0iSES5FaINMvgWe-JRoiRSmg4cdGS-KbU,189
+diffeval/backends/diffusers/attention_name.py,sha256=PZ5M_bs-qvTnk5nI1KBO-7NHmjyWtQ8NQnTaoMuUN_c,447
+diffeval/backends/diffusers/attention_sdpa.py,sha256=gdcr0XLOZ72GCzf1Twqr8xNtgaBYpTEFsn4GtjhJqgE,187
+diffeval/backends/diffusers/attention_xformers.py,sha256=ZkpdvHCXviTpOtY0zlZg7L6fM4FHbCwfVtUT-i1LMM0,128
+diffeval/backends/diffusers/backend.py,sha256=OVgoPHhyFcToaJawsSKBL_McDgbz_1d_CSPtRD7LvF4,1122
+diffeval/backends/diffusers/call_kwargs.py,sha256=fLXwzXLewhJjXSdLk13UFtS_jWSRSjce1M7fcrmCa0g,685
+diffeval/backends/diffusers/coerce_dtype.py,sha256=AuAQepjrkV5QmmSUOUXIv4DN-oJqDzJf17WYTwnKIYQ,348
+diffeval/backends/diffusers/describe.py,sha256=tpbKQ8OacrkVZs-SjdlH7Jq6hBmeb3jKN1QMjgF7YIM,697
+diffeval/backends/diffusers/dtype_name.py,sha256=6Pll7Hib5h_lrygRUTn2YP9CVT-JQHg2DjBuNZrsSLg,703
+diffeval/backends/diffusers/from_config.py,sha256=pDdpTDCrKF_nshJBr7eLYK6Q7nwxVR_0-4VAE-aCYIo,788
+diffeval/backends/diffusers/generate.py,sha256=-QGtrUXMFSZDmwbhxZ91sMvUQHHVrOdPHz35tsB0eMI,706
+diffeval/backends/diffusers/kwargs_generator.py,sha256=DEdBNeVi6R2nzOG94Rq--D5Qgln1PsIu-QzpXFk80TM,166
+diffeval/backends/diffusers/kwargs_prompt.py,sha256=VtzwmRkfeulfmGgZZnI3gW6mxEVaOHCPv1q92R5iFok,441
+diffeval/backends/diffusers/kwargs_sampling.py,sha256=7waAXytyIQVXf4imWj8kaVOKnUkMpbKe4NkcHJPxwOU,618
+diffeval/backends/diffusers/kwargs_size.py,sha256=TunIUieZp9YB2Ij93aOoNxAHAVJg4sd441aa_dqRARs,517
+diffeval/backends/diffusers/load_kwargs.py,sha256=THFBjE8CoI3UltF8-x6c-uh5Ue3yYNDaHmAWqjKU1XE,265
+diffeval/backends/diffusers/load_pipe.py,sha256=-ZWAoSuxvJYsykJt2KpNKBmliJjkX9oj0MPOiVTh-9c,1237
+diffeval/backends/diffusers/missing_dep.py,sha256=_pQlC7a-YcEumeVZi6uUsVULvGZfv0DwjE4UsNseNws,312
+diffeval/backends/diffusers/one_batch.py,sha256=sYQZoNzf3n3z81bL7w6dDBmcoohum8CMGFQqhCoCqJs,465
+diffeval/backends/diffusers/pick_device.py,sha256=g0HTwRLSyU2Ryc5A8eb7e6k8B6MG2OvT7WK2dgLU-MI,340
+diffeval/backends/diffusers/presets.py,sha256=cNB-BGCWwoXwYH_jSY0YdXBv7LV8hwOvwvgAg_ID_1g,231
+diffeval/backends/diffusers/sampling.py,sha256=J4doyi1-C44U7c4miRI6n6PcrHBdMOfXgm0Ke8w-ejM,707
+diffeval/backends/dummy/__init__.py,sha256=8rDTezeDWf81JdKeCWCUGm-GrMdKQ_niRk7tvyE0xc0,47
+diffeval/backends/dummy/backend.py,sha256=T_EAsuSnDYmUXK0XZ9UMwB3fu4sKR64OOBiN0Ylsmso,950
+diffeval/backends/dummy/box.py,sha256=uAyiQ5-amr1kzfwnEq35agSxI6rYvBFLj7TjIOwWZz8,346
+diffeval/backends/dummy/colors.py,sha256=uzwAG5k0gILoI6KM8e0nReOGTMuKdey0UTlMx69ImTE,369
+diffeval/backends/dummy/describe.py,sha256=UeXxA_y_N3pRk6tyosvr_zYsljz01axiQxgMggfhBxM,591
+diffeval/backends/dummy/draw.py,sha256=ayXxcP9GiwKvNKsM8ozpTDfbbIZmrOtRUHnxNvmnvAk,476
+diffeval/backends/dummy/from_config.py,sha256=RxyY3CqbUrANcozqkgX1ngZD6TAhdVMIcPIHWS4n7h0,570
+diffeval/backends/dummy/generate.py,sha256=6Utx4qRWMXnaInv6wkZJ_Dfp3gfUxsMeg1CalawsPY0,642
+diffeval/backends/dummy/render.py,sha256=F5Zd36SXFr298d1_n7JfZrp770WQ1cuWFs9VfFBtwoo,613
+diffeval/bench/__init__.py,sha256=Y5RFNKiPHgvKBVuPRbaUuenhBmgj-0tgF6JL1jFdbb0,85
+diffeval/bench/cmmd_check.py,sha256=UC-DEzYuPfxsNNQ6F4066il1LRoAkKxkIRIFR4t_GJg,795
+diffeval/bench/cmmd_samples.py,sha256=ZRETpulwZmHZ436tNq0QYAWGmCYwX2TdkD6YN5TszMM,612
+diffeval/bench/coco30k_cli.py,sha256=O7VDfuXba08MllrKDtXbbvo38CqeqmOG_EjLpqgrEsE,1189
+diffeval/bench/coco30k_export.py,sha256=qkFIr8pDVMdem4Orw2tdEZaPe7sTnygH9YuA7jXCnXE,743
+diffeval/bench/coco30k_files.py,sha256=eVqp05hiR6iWUvoAaDEKdDHKZM-WHXGm5bUiy0P3pSU,691
+diffeval/bench/coco30k_images.py,sha256=umdf2S3ttlk_JJrvdQW8w25qCjJVEXYY7wyAUgbufdo,765
+diffeval/bench/coco30k_suite.py,sha256=HjHi8gO_zCG6eLl0g2ZvWs9qNPY6hWJnDe62-gy1KKY,799
+diffeval/bench/fetch.py,sha256=mUiNEhPtU42DWokO86cJX94OM_VgHjUP-i3NG8RPtuI,325
+diffeval/bench/published.py,sha256=2LwI1ZjTZ24EwdT6B5btTZ1BPiQM6DqIblgBU-f6Pyg,397
+diffeval/bench/table_cli.py,sha256=qixXR8iULjof6teyI4uVq1bARYIUu8Im585eFo93T_U,488
+diffeval/bench/table_row.py,sha256=EHE7TEHkl3TewLhbACqQs4gQMuWRPExADCH6ADjvxuU,1081
+diffeval/cli/__init__.py,sha256=4V0irZMNErUIhrVB5uA1WagP4ILj3e-vUTRUKux4XqM,31
+diffeval/cli/command_list.py,sha256=kTXxWBwhCUFgz5afpla4eqqNahe0B27N2dFjx5u9F3o,379
+diffeval/cli/command_run.py,sha256=wQbpbg3ZKSfxDldkw9V5A1EEhylbwAwE9ikm2hZ6HP8,977
+diffeval/cli/config.py,sha256=KQboxs9LGsf4bp9BJ_m1DpEnolCo_A9ZPY8KubXlDQM,594
+diffeval/cli/csv_ints.py,sha256=xvY3-dOFLFORBh8h2p-XiWPaMRXfqCc30g8txgL_pEg,352
+diffeval/cli/csv_names.py,sha256=fQczfuFffOiXn-fqmfkSplLI2zcxnDZYizD11edAWhI,266
+diffeval/cli/main.py,sha256=rTDmoSY6P2LOZ75cZcL_ArHRH9OxHatK2srJUNFPpUw,403
+diffeval/cli/output_dir.py,sha256=HFw9F5oY2bauVTvQWVdUVUrmQeRi7HkorQObLVaw8ds,432
+diffeval/cli/parser.py,sha256=lXaFup2gji_D7NWX6q2MhE5IUJrPxo_thecHDPPfxAY,572
+diffeval/cli/print_paths.py,sha256=FatkHa_VShGa7t4I-GTUm5f21k8iN49SgijMG9DYeTs,420
+diffeval/cli/print_registered.py,sha256=H0535bO0RRGlg-IyUAUKUcNn7saLGuag8D5ceGIMDbk,216
+diffeval/cli/run_parser_eval.py,sha256=7sGCV43aDgRqX9Czk8ImbqdqLgReVtsK2WfTUnswaTA,589
+diffeval/cli/run_parser_model.py,sha256=kkdhsr_ezSkxy0CtHdjmp8xQ6iAxAgN__jMfHIb76G8,639
+diffeval/metrics/__init__.py,sha256=b4OClYTmjGI1UxZ58bhmdE0nIzFllHbqAft4AjbPu8g,228
+diffeval/metrics/count_warning.py,sha256=5XaGrq_Y0ko7mtZjJJQYAyUlLF_rjjno8pMEtZlpKcg,253
+diffeval/metrics/images.py,sha256=hW9MAGM_QesQvVmx1yC-fVlgpdbcCYlQ2Z3VAy2opEo,329
+diffeval/metrics/ok.py,sha256=CXCrbRCd8gJJZvW2C29jL9YbpXfka_UYESpwnz4_oQ4,360
+diffeval/metrics/protocol.py,sha256=Ptmbw2NQTfHC6TKayxGXNebscnDLBpZvS0qX8Sl_yoo,438
+diffeval/metrics/skipped.py,sha256=6rlzM-t_kTsN2CfIlbJKFWPjD_kVtr9NGYGmuQjOPxY,415
+diffeval/metrics/clip/__init__.py,sha256=DkA5nOCJNqKeDaQzRIb-l3_zQ7Y218l8fAH1PF-GNbE,25
+diffeval/metrics/clip/features.py,sha256=iNaZqdA_MtiSL5bfXQ0OysGi2XdNiT_JpEtNuWRZ22c,481
+diffeval/metrics/clip/formula.py,sha256=QO6zNPCzaQ0HBWf0xChXkpKg_-eIwjfhyekRfZ4-6xg,201
+diffeval/metrics/clip/load.py,sha256=GVhZlEIb_IL0UlhyU6zCryGDGwr5GcRcrgFAyvCZbKw,477
+diffeval/metrics/clip/metric.py,sha256=9YYNOO98jFthBlcxKzlPLA48s2eb7mMLKz8N1bi0MLM,591
+diffeval/metrics/clip/protocol.py,sha256=g0Q_8_Mo_O0cDDEVW7qZagzrsQjj1FPRniBxNaiOGco,318
+diffeval/metrics/clip/sample_image.py,sha256=UOqi1wTJupoBETG3PIsLjAkUsGPzFBMXVqNVHdOz9Nk,331
+diffeval/metrics/clip/score.py,sha256=a5_gDN1KYe5Um6KD6QVt6TDyvXH-oOViJwANzepB1wo,1091
+diffeval/metrics/cmmd/__init__.py,sha256=Xpi1fLyAR5OTkE5-Fm3VBPd9DM0lJsnNZTAntIv2NaY,20
+diffeval/metrics/cmmd/embed_batch.py,sha256=IWLdXwyZEGSiUzyyBttUPMUNMqE8OEq3wFdDealop9s,968
+diffeval/metrics/cmmd/embed_paths.py,sha256=UJA-NhjDmtFDx3feAZNgJdK_NEsVQwBjnwqTTsqiFSs,1055
+diffeval/metrics/cmmd/metric.py,sha256=ygnkUjttqAkV9UBar-A57Yo6lTCRzuPvlZIrqHfdvb0,559
+diffeval/metrics/cmmd/mmd.py,sha256=MrhyyN0WjbssLiSQEl-fauMHUJCZcYxpKjF5RwRkA7Q,783
+diffeval/metrics/cmmd/protocol.py,sha256=QyBzdNW5CrAoKlPS9cDr15xjJc1jNwItrbqfEXcF6yc,457
+diffeval/metrics/cmmd/resize.py,sha256=0QNnIZ3KLC_C43Ba_Bzw1f7J6L9U60V9v09ygTwQl3Q,540
+diffeval/metrics/cmmd/score.py,sha256=JYUxF6jZbLgo6yr41NEeYhzJeV3898_AojpKlSoOKFw,1156
+diffeval/metrics/distribution/__init__.py,sha256=DIKIRdg4pVnimFZR11myPIR1UO5o0zY5cBi0Hpig5lE,20
+diffeval/metrics/distribution/call.py,sha256=aFDBb02Duwm3gLcKLtQ2gecfkM6N3pURFaNXDoGs-3c,1113
+diffeval/metrics/distribution/kwargs_folder.py,sha256=hoJTX5129a72IoaxBUjyRhoT6Zt3h8WjTSptzB-dtYQ,829
+diffeval/metrics/distribution/kwargs_stats.py,sha256=V6_O8DJFEbdzdlzzkgmJQ7mb5TqNehIEP6uwLbmV8M4,502
+diffeval/metrics/distribution/protocol.py,sha256=16wpT4Ov6d00buKXJ0U_AoJA9_9QhVQJNXUIxujVOj0,480
+diffeval/metrics/distribution/seed_numpy.py,sha256=2TFW3qHFqxx9dYbQssmqNiFE51N9K2ZonZkhupevylE,255
+diffeval/metrics/fid/__init__.py,sha256=5CwMV1VE5wZzGHv5nN74T0_IgEFS-_l9NZ7nssJN7Nk,19
+diffeval/metrics/fid/guard.py,sha256=q_AZ1mV0JjiiY5rEtKLGroYuDocaHlJ-pcZt_cwoZwM,533
+diffeval/metrics/fid/metric.py,sha256=IshgH2zdNSO469Rd0gW-3-pg27tfTMiCoGp1u3s9QRk,594
+diffeval/metrics/geneval/__init__.py,sha256=sFAQiXzXV02ANdAhG6g1J28wUoYKxOYpZTjfZCJK39w,21
+diffeval/metrics/geneval/metric.py,sha256=HK9AdDSgMrIALAA7lKXBucMdpreOMAQ_1z11KcrVPtk,744
+diffeval/metrics/geneval/protocol.py,sha256=eXFMbh8eEpMjTNFhhSFs4o7tKGK_DzYU63VNaKD59bU,165
+diffeval/metrics/geneval/tags.py,sha256=6KlzO3_9esYy2DTXoorDJrVNKHu3MhyFXCSZ2yFEUUU,124
+diffeval/metrics/kid/__init__.py,sha256=Nn8wpgHxbudKyDZPG6sPYPu-N9lmNH-uHDF93cNYe5c,19
+diffeval/metrics/kid/guard.py,sha256=-Ivtp9qZd3SNnuUuMD-CjQf6P0mBad7hy40Btg1MciU,111
+diffeval/metrics/kid/metric.py,sha256=rTluSVLFQGQoZ6CD2DA1mAtoB8xHby58A29ZhrBd7PM,585
+diffeval/registry/__init__.py,sha256=InO7Er1SjXtwIBasKKNb-cLGB7ZTdyVoLq7qGcV1iCk,45
+diffeval/registry/lookup.py,sha256=iM8L-3l3z5hpI6S4xxKX2NwX-i028HgfdpTd-KQ5sMQ,285
+diffeval/registry/names.py,sha256=JDu7NCIdCNFFhvSdN7B8EYx276zLMCmsLbG2JCjFClA,98
+diffeval/registry/register.py,sha256=frqKb2lQbb2kEQfUu1A4EVY4eToqpUdFxNW4mHggT-g,299
+diffeval/registry/tables.py,sha256=Q-McMkR9t_FEFe18jlrMOYTygCWBZbRQMl7pxr90sG8,113
+diffeval/registry/unregister.py,sha256=Iz9LfKGMBVjgHpUDf-VEEyPwesCma0qR1JWs7qyo2wU,112
+diffeval/report/__init__.py,sha256=ihoQNZFKSnC5JL7Q18qEjZZEFoD4KPF9DScRPLWYKqo,56
+diffeval/report/console.py,sha256=PJU72cHiOut8P0EbJ4z3UJm72X8MUDRigUcEz7TktDY,768
+diffeval/report/console_widths.py,sha256=bPYaMF9nmEZc_Hwlj6C9BgHn2fbO8FXmOI74hg0nqmY,178
+diffeval/report/direction.py,sha256=VslsJLAPBkyzhxUSPIlru6GURN1yfMhZu48CPp1Pfwg,198
+diffeval/report/html_figure.py,sha256=8NbmDHIMATKlHOJtEmVSLkNNUxLOV75sjv22WqKVBX0,300
+diffeval/report/html_grid.py,sha256=2pmRmpxPmQPtrVPgSmoXnP5SC7qNGnWUnrOozdgYXaI,662
+diffeval/report/html_page.py,sha256=gB9wRigrh2WaikqtHKG7ZR9Z4GB5HCSqj0robz2S51o,585
+diffeval/report/html_row.py,sha256=LkGDLuVg2ZUEuBgtsUfS27YLfDZXS2NLmgjPnGXr_1k,477
+diffeval/report/markdown.py,sha256=ID51MIEZsy_TTecp0KUFs1ZyeXGGaB8fDyO82IKEIBw,445
+diffeval/report/markdown_row.py,sha256=LDMlcAibs2k8VT1B2rVC8PEo_QlSUO0lEUQ5XxTqlek,409
+diffeval/report/number.py,sha256=NcO0BDscEyM8ui_jMwt_w3FzMwt-LXhWRvvpym8DPmo,167
+diffeval/report/templates/page.html,sha256=4piZfB7n-Z9PGuKjfLuM0K0mEQcLyucTgE5FTasQz9I,1191
+diffeval/runner/__init__.py,sha256=46wQOnD1V-OP8z3_VGQchV_-5eVbx8wzpQ2t2oiZrBQ,60
+diffeval/runner/generate_save.py,sha256=MuS44klW3dhjjUl82QPXMJg1Nl2IxqMRoxtdROkrP5U,660
+diffeval/runner/metric_device.py,sha256=40f9l73sBAKqCcfmHyRCjbUZR1hrScRqFqG8QjMfyYA,408
+diffeval/runner/one_seed.py,sha256=_hPV4QwIEVPe0BzV6rwRiRU6fikalA0zzXKiT6TrNBo,1109
+diffeval/runner/options.py,sha256=gnRmIWxBrPaByzvsrETRw1_zEQI7pU013CPK3AJk__U,662
+diffeval/runner/payload.py,sha256=UA-VdDlBmllG-uxvliuaZ4D8cNK26dQVHiiRokGWLkA,1435
+diffeval/runner/relpath.py,sha256=BK45uqfUplwOWQ3e7nyEY5OiJeLFmBtkWVwlJS4YFLw,187
+diffeval/runner/requests.py,sha256=q-CqJ34dQYWqDbHizzsCiq7TgHnCdMY8T4YslRlbYGs,294
+diffeval/runner/run.py,sha256=cGGeq_zBhj8tjphoVwDgusJw6awKWOqV59zJiRn_6ds,963
+diffeval/runner/safe_metric.py,sha256=xh4oy4ZZD5AEtBMDdwPyjEwZCRAqCbU9qrDyHiAc2PE,737
+diffeval/runner/save_png.py,sha256=jR66uqBgb_TZ78168GqTPzznVJU5_65bX58bG0T5JRY,209
+diffeval/runner/seeds.py,sha256=KPre4lcec9oj12leD7at_tZa6fBQEIN0wIHSAtPL2ok,875
+diffeval/runner/write_json.py,sha256=TcD1zkGwCWYKhZwMqj6vdMlzIHLQlN7WwMXdiYOHb24,316
+diffeval/runner/write_reports.py,sha256=_gZLzXNs2VDqQvSjIVZ9JpQoil6Bs6imPOiPqWBVxHs,487
+diffeval/system/__init__.py,sha256=MtgwEpd10OuJAuKtZUgjbxFJ6EIJirnrE_WsiDXRJ-g,53
+diffeval/system/describe.py,sha256=wiqwMrrM8yBjVWu7wfZ4gFyhrdizxud6ncGNQApNF7U,271
+diffeval/system/gpu.py,sha256=vj-vqvufFvt6TlxNIB5YmWxk71hpdYfVBYL6mxchnZk,284
+diffeval/system/gpu_cuda.py,sha256=Y4T9UFHPrv7bkG0JwB4QAeFwIbHS78y8usCUjZLb5lw,352
+diffeval/system/gpu_mps.py,sha256=ZS1CJ-P173gX7U7klrRuGkRnoOgou9VlRNZ2vD0HDOU,268
+diffeval/system/sdp_flags.py,sha256=nsOY59bQ4X4K8vyrbogLC2-pQK8kp9Q3U1EAvRfiOm0,410
+diffeval/system/version_of.py,sha256=59-E_HnFSycOtysWXZ8jRhDKrBE46ISYFKcY6Iuecqg,248
+diffeval/system/versions.py,sha256=iTgRPPJrgpZHQlz8XAf2zwV15aDRLJtwLfVf0flpSDI,619
+diffeval/tasks/__init__.py,sha256=0muiLn5rushiv3oJm33VA823cYoMdfkhiT8p-pFTC7U,66
+diffeval/tasks/build_suite.py,sha256=lPtxGLniqYoS91rzGp8dJjvF5aYmW-ngNy0zU_KFnNU,159
+diffeval/tasks/core.py,sha256=nmClOKMQxEbw1aHcZiJgzo4_Xky0n_40m48K8nVUuiU,373
+diffeval/tasks/errors.py,sha256=gqqet21orgb8iycWlHknFYapJou_2yr0AOCn6j2Tg0Y,140
+diffeval/tasks/fields_metrics.py,sha256=uRivVz12QpmTPXyoOHCevc8gxMocD1vJadupm175XnA,489
+diffeval/tasks/fields_paths.py,sha256=13ZJIijaOxP3aekNgAervi_z14BNHA9h8sfYkt3j4hU,801
+diffeval/tasks/fields_prompts.py,sha256=6t7Qbs6izZNDVaXvStM42IvnpJP3A5gfDdrMUFLKqE4,588
+diffeval/tasks/fields_seeds.py,sha256=uDgp84BNYPmffEGTpEo9qLB6OCkXfIlWNOC0mbjFa4M,488
+diffeval/tasks/fields_size.py,sha256=T8SumjGDrpIKqA7l68GZHpz9p2G3IemxAdQXd3lt53s,429
+diffeval/tasks/fields_text.py,sha256=1OLkircSqSSGDsGfLoRdZSj4UOHJYC6mBEcBmN8le3k,473
+diffeval/tasks/from_mapping.py,sha256=hw5FKG9OqbhbZ4sf2HsJEz8zYVNHyEtbrPN2_WkmCjo,786
+diffeval/tasks/load_bundled.py,sha256=yhXOlgWMm_LCW8RxKaI8VIUwsECsD9e27PFwvgWtizQ,225
+diffeval/tasks/load_path.py,sha256=VT5gv1TrZ4005sZAeKEsr3kgcfwZYqQIxWHanjC2VXc,345
+diffeval/tasks/one_prompt.py,sha256=yZS4nBlmN9YaDaVMa_fkNIS7AOLCmwHoU5pTmoDqu_4,682
+diffeval/tasks/prompt_id.py,sha256=dTXrQTMFCPdhADnLgQVtEQF5LyPEo2AqpWwviEcQRKo,321
+diffeval/tasks/read_yaml.py,sha256=7X-nEdmwhFKkXOtRqf_YFxUgM7jDKwjT_IofYl64-DY,344
+diffeval/tasks/resolve.py,sha256=wruFXRgmnyW-2cc8B3Y4udRuaAw--I9Fk2DYYe9Kti8,532
+diffeval/tasks/suites/core.yaml,sha256=4lTKfD2EAr9AgLFGP9iT2PSoBOVAfjzzjrWn5b-fTwQ,375
+diffeval/tasks/suites/references/core/blue_sphere.png,sha256=HXEojB2OGFxJp3EpdMQbhZ12keR4UZcKOTfbRv7lFGs,329
+diffeval/tasks/suites/references/core/green_triangle.png,sha256=9lOmS1eEURUYpBKIzP9dzw5fgb99QbwUiWioge9V-94,294
+diffeval/tasks/suites/references/core/red_cube.png,sha256=WKJT2u5KmCuRS6hBLtl4qcDnAC1XQgBGkTeQ6HVkrDc,187
+diffeval/tasks/suites/references/core/yellow_star.png,sha256=rCrK1xlMS51mwP3E5MiFgDdaKttHtxoRTOtNx3ZQix8,397
+diffeval/types/__init__.py,sha256=HIwvdy9hKN_v6wi2jBSTj2Rfpo_SE_fWt3VC2kUaEJ8,55
+diffeval/types/context.py,sha256=Hn2MHmsB8dZmmsVmH9EWxoojGQo4ZX6mW6fZYEKYLHw,331
+diffeval/types/prompt.py,sha256=OwG7zmgQfqcnZxEhkHgc47ZQ5xFBHxFYCVnikidJ0jg,181
+diffeval/types/request.py,sha256=cBr8q-ehdgtD-TrdB6y_kB4_1Ob6_Vznm66o5OLXeHA,266
+diffeval/types/result.py,sha256=HRrYqQrnLIDPIOTr5kq4_fZ_c5bm0VGlNXIUyYERb5M,266
+diffeval/types/result_rules.py,sha256=khgbvkaOoTMh5bHEMIoLGtDQdXOF-pmaYPxqQOWa9bM,487
+diffeval/types/sample.py,sha256=Ri5OFIqML6J33KuAU1F7N3ZJv9mjRpmHXx0qWA67M5Q,302
+diffeval/types/suite.py,sha256=mtJq7x4xmqw84TKMTVtTIRzqFlEk-SqYZFOlVgLv9EU,466
+diffeval-0.1.0.dist-info/licenses/LICENSE,sha256=6iwF4Dfzd9-RiXgnsNEc3Y8kjQL_9hxmg6lWYRIt5HE,10942
+diffeval-0.1.0.dist-info/METADATA,sha256=Si2bAhdJLkr6w4twgaeq0e_DIr6kMo-hv0qJqEe7yBs,19237
+diffeval-0.1.0.dist-info/WHEEL,sha256=mOdXR9C3wMVov7cSY8PY-XCzo3SmmWCHvGQr5csJoao,91
+diffeval-0.1.0.dist-info/entry_points.txt,sha256=0nR7fAYKY0kCRbj4dkstDwH0bE10RmKcm7rx4Jaw6-k,264
+diffeval-0.1.0.dist-info/top_level.txt,sha256=JELkY7mzlg3cjdeYySbydrAXy1f1icqx9mi6RjD0xOY,9
+diffeval-0.1.0.dist-info/RECORD,,

diffeval-0.1.0.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (84.2.1)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

diffeval-0.1.0.dist-info/entry_points.txt

@@ -0,0 +1,6 @@
+[console_scripts]
+diffeval = diffeval.cli.main:main
+diffeval-check-cmmd = diffeval.bench.cmmd_check:main
+diffeval-coco30k = diffeval.bench.coco30k_cli:main
+diffeval-coco30k-images = diffeval.bench.coco30k_images:main
+diffeval-table = diffeval.bench.table_cli:main