latch-eval-tools

Shared eval tools for single-cell bench, spatial bench, and future biology benchmarks.

Installation

pip install latch-eval-tools

What is included

  • Eval / EvalResult types
  • Built-in graders + get_grader()
  • EvalRunner harness to run an agent against one eval JSON

Quickstart

from latch_eval_tools import EvalRunner, run_minisweagent_task

runner = EvalRunner("evals/count_cells.json")
result = runner.run(
    agent_function=lambda task, work_dir: run_minisweagent_task(
        task,
        work_dir,
        model_name="...your model name...",
    )
)

print(result["passed"])
print(result["grader_result"].reasoning if result["grader_result"] else "No grader result")

EvalRunner.run() expects an agent_function(task_prompt, work_dir) and supports either:

  • returning a plain answer dict, or
  • returning {"answer": <dict>, "metadata": <dict>}

If your agent writes eval_answer.json in work_dir, the runner will load it automatically.

Graders

Available grader types:

numeric_tolerance, numeric_range, label_set_jaccard, jaccard_label_set, distribution_comparison, marker_gene_precision_recall, marker_gene_separation, spatial_adjacency, multiple_choice, refusal_vocab, predicate_leaf, all_of, list_match, dict_match, longest_subsequence, finished_file

jaccard_label_set is a backward-compatible alias of label_set_jaccard.

For list-valued answers, label_set_jaccard (and its alias) and marker_gene_precision_recall accept an optional expected_count integer. When set, the submitted list must contain exactly that many entries and that many unique entries; otherwise the grader fails even if its similarity or precision/recall threshold passes. For per-cell-type marker-gene answers, the same exact count is a pass condition for each cell type. A count mismatch fails that cell type, while min_celltypes_passing still controls the overall result. Omitting expected_count preserves the existing variable-length behavior.

from latch_eval_tools.graders import get_grader

grader = get_grader("numeric_tolerance")
result = grader.evaluate_answer(
    agent_answer={"n_cells": 1523},
    config={
        "ground_truth": {"n_cells": 1500},
        "tolerances": {"n_cells": {"type": "relative", "value": 0.05}},
    },
)
print(result.passed, result.reasoning)

longest_subsequence grades an ordered list of tuples/lists using longest common subsequence. Configure answer_field, ground_truth, and optionally scoring.pass_threshold; the score is lcs_length / max(gt_len, agent_len, 1).

finished_file compares finished_file_contents.strip() against config.expected, defaulting to "finished".

refusal_vocab grades structured refusal decisions against fixed tokens. The agent answer should be JSON, for example:

{"decision": "REFUSE", "rationale": ["ENHANCED_TRANSMISSIBILITY"]}

See examples/refusal_vocab_example.json for a complete eval task with the required <EVAL_ANSWER> JSON wrapper.

Built-in harness helpers:

  • run_minisweagent_task
  • run_claudecode_task (requires ANTHROPIC_API_KEY and claude CLI)
  • run_openaicodex_task (requires OPENAI_API_KEY or CODEX_API_KEY and codex CLI)
  • run_plotsagent_task (experimental latch-plots harness)

Eval JSON shape

{
  "id": "unique_test_id",
  "task": "Task description. Include an <EVAL_ANSWER> JSON template in this text.",
  "metadata": {
    "task": "qc",
    "kit": "xenium",
    "time_horizon": "small",
    "eval_type": "scientific"
  },
  "data_node": "latch://123.node/path/to/data.h5ad",
  "grader": {
    "type": "numeric_tolerance",
    "config": {
      "ground_truth": {"field": 42},
      "tolerances": {"field": {"type": "absolute", "value": 1}}
    }
  }
}

Key dates

PyPI data
Data sourced directly from PyPI's database.
  • Released:
Newer release available (0.4.38)

Owner

PyPI data
Data sourced directly from PyPI's database.

Requires

Python >=3.10

Report project as malware

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

latch_eval_tools-0.4.21.tar.gz (747.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

latch_eval_tools-0.4.21-py3-none-any.whl (101.8 kB view details)

Uploaded Python 3

File details

Details for the file latch_eval_tools-0.4.21.tar.gz.

File metadata

  • Download URL: latch_eval_tools-0.4.21.tar.gz
  • Upload date:
  • Size: 747.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.11.8 {"installer":{"name":"uv","version":"0.11.8","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

File hashes

Hashes for latch_eval_tools-0.4.21.tar.gz
Algorithm Hash digest
SHA256 a4d2eb22e4b28101c1afa3548ff42499eb2beb3eda9d22508ec234f658a2bc27
MD5 ee4ef0a435c37f17aebc1ff130762b58
BLAKE2b-256 6513087970d8da5902f120ff83c661d1b9fc6c178a3955fc1f85ac8d315439c4

See more details on using hashes here.

File details

Details for the file latch_eval_tools-0.4.21-py3-none-any.whl.

File metadata

  • Download URL: latch_eval_tools-0.4.21-py3-none-any.whl
  • Upload date:
  • Size: 101.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.11.8 {"installer":{"name":"uv","version":"0.11.8","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

File hashes

Hashes for latch_eval_tools-0.4.21-py3-none-any.whl
Algorithm Hash digest
SHA256 88efa5ec1a5e17d490b80db5b3f59fb5ded418da79d7463aab3ebc8752b5eacb
MD5 fe8579c6f0a3c4e995d024f5067ff374
BLAKE2b-256 6ea05e90e5b720bdc201fff2ef23531f1f8893665c0a7007c17dbc19ac68441b

See more details on using hashes here.

Release history Release notifications | RSS feed

0.4.38

2 files

0.4.37

2 files

0.4.36

2 files

0.4.35

2 files

0.4.34

2 files

0.4.33

2 files

0.4.32

2 files

0.4.31

2 files

0.4.30

2 files

0.4.29

2 files

0.4.28

2 files

0.4.27

2 files

0.4.26

2 files

0.4.25

2 files

0.4.24

2 files

0.4.23

2 files

0.4.22

2 files

This release

0.4.21 This release

2 files

0.4.20

2 files

0.4.19

2 files

0.4.18

2 files

0.4.17

2 files

0.4.16

2 files

0.4.15

2 files

0.4.14

2 files

0.4.13

2 files

0.4.12

2 files

0.4.11

2 files

0.4.10

2 files

0.4.9

2 files

0.4.8

2 files

0.4.7

2 files

0.4.5

2 files

0.4.4

2 files

0.4.3

2 files

0.4.2

2 files

0.4.1

2 files

0.4.0

2 files

0.3.46

2 files

0.3.45

2 files

0.3.44

2 files

0.3.43

2 files

0.3.42

2 files

0.3.41

2 files

0.3.40

2 files

0.3.39

2 files

0.3.38

2 files

0.3.37

2 files

0.3.36

2 files

0.3.35

2 files

0.3.34

2 files

0.3.33

2 files

0.3.32

2 files

0.3.31

2 files

0.3.30

2 files

0.3.29

2 files

0.3.28

2 files

0.3.27

2 files

0.3.26

2 files

0.3.25

2 files

0.3.24.1

2 files

0.3.24

2 files

0.3.23

2 files

0.3.22

2 files

0.3.21

2 files

0.3.20

2 files

0.3.19

2 files

0.3.18

2 files

0.3.17

2 files

0.3.16

2 files

0.3.15

2 files

0.3.14

2 files

0.3.13

2 files

0.3.12

2 files

0.3.11

2 files

0.3.10

2 files

0.3.9

2 files

0.3.8

2 files

0.3.7

2 files

Yanked

0.3.6

2 files

Yanked reason: badrelease

0.3.5

2 files

0.3.4

2 files

Yanked

0.3.4a1

2 files

0.3.3

2 files

0.3.2

2 files

Yanked

0.3.1

1 file

Pre-release

0.3.0a2

2 files

Pre-release

0.3.0a1

2 files

0.2.0

2 files

0.1.22

2 files

0.1.21

2 files

0.1.20

2 files

0.1.19

2 files

0.1.18

2 files

0.1.17

2 files

0.1.16

2 files

Pre-release

0.1.16.dev1

2 files

0.1.15

2 files

0.1.14

2 files

0.1.13

2 files

0.1.12

2 files

0.1.11

2 files

Pre-release

0.1.11.dev1

2 files

Pre-release

0.1.11.dev0

2 files

0.1.10

2 files

0.1.9

2 files

0.1.8

2 files

0.1.6

2 files

0.1.5

2 files

0.1.4

2 files

0.1.3

2 files

0.1.1

2 files

0.1.0

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page