From 093832c2c290ff9f6f68a2f87964b78402539deb Mon Sep 17 00:00:00 2001 From: phranzia <9207473+phranzia@users.noreply.github.com> Date: Wed, 12 Aug 2026 15:25:50 -0700 Subject: [PATCH] edits --- LICENSE | 366 +++++++++++++++++++++++++++--------------------------- README.md | 3 +- SKILL.md | 144 ++++++++++++--------- 3 files changed, 271 insertions(+), 242 deletions(-) diff --git a/LICENSE b/LICENSE index 37bcea5..89471bc 100644 --- a/LICENSE +++ b/LICENSE @@ -2,180 +2,180 @@ Version 2.0, January 2004 http://www.apache.org/licenses/ - TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION - - 1. Definitions. - - "License" shall mean the terms and conditions for use, reproduction, - and distribution as defined by Sections 1 through 9 of this document. - - "Licensor" shall mean the copyright owner or entity authorized by - the copyright owner that is granting the License. - - "Legal Entity" shall mean the union of the acting entity and all - other entities that control, are controlled by, or are under common - control with that entity. For the purposes of this definition, - "control" means (i) the power, direct or indirect, to cause the - direction or management of such entity, whether by contract or - otherwise, or (ii) ownership of fifty percent (50%) or more of the - outstanding shares, or (iii) beneficial ownership of such entity. - - "You" (or "Your") shall mean an individual or Legal Entity - exercising permissions granted by this License. - - "Source" form shall mean the preferred form for making modifications, - including but not limited to software source code, documentation - source, and configuration files. - - "Object" form shall mean any form resulting from mechanical - transformation or translation of a Source form, including but - not limited to compiled object code, generated documentation, - and conversions to other media types. - - "Work" shall mean the work of authorship, whether in Source or - Object form, made available under the License, as indicated by a - copyright notice that is included in or attached to the work - (an example is provided in the Appendix below). - - "Derivative Works" shall mean any work, whether in Source or Object - form, that is based on (or derived from) the Work and for which the - editorial revisions, annotations, elaborations, or other modifications - represent, as a whole, an original work of authorship. For the purposes - of this License, Derivative Works shall not include works that remain - separable from, or merely link (or bind by name) to the interfaces of, - the Work and Derivative Works thereof. - - "Contribution" shall mean any work of authorship, including - the original version of the Work and any modifications or additions - to that Work or Derivative Works thereof, that is intentionally - submitted to Licensor for inclusion in the Work by the copyright owner - or by an individual or Legal Entity authorized to submit on behalf of - the copyright owner. For the purposes of this definition, "submitted" - means any form of electronic, verbal, or written communication sent - to the Licensor or its representatives, including but not limited to - communication on electronic mailing lists, source code control systems, - and issue tracking systems that are managed by, or on behalf of, the - Licensor for the purpose of discussing and improving the Work, but - excluding communication that is conspicuously marked or otherwise - designated in writing by the copyright owner as "Not a Contribution." - - "Contributor" shall mean Licensor and any individual or Legal Entity - on behalf of whom a Contribution has been received by Licensor and - subsequently incorporated within the Work. - - 2. Grant of Copyright License. Subject to the terms and conditions of - this License, each Contributor hereby grants to You a perpetual, - worldwide, non-exclusive, no-charge, royalty-free, irrevocable - copyright license to reproduce, prepare Derivative Works of, - publicly display, publicly perform, sublicense, and distribute the - Work and such Derivative Works in Source or Object form. - - 3. Grant of Patent License. Subject to the terms and conditions of - this License, each Contributor hereby grants to You a perpetual, - worldwide, non-exclusive, no-charge, royalty-free, irrevocable - (except as stated in this section) patent license to make, have made, - use, offer to sell, sell, import, and otherwise transfer the Work, - where such license applies only to those patent claims licensable - by such Contributor that are necessarily infringed by their - Contribution(s) alone or by combination of their Contribution(s) - with the Work to which such Contribution(s) was submitted. If You - institute patent litigation against any entity (including a - cross-claim or counterclaim in a lawsuit) alleging that the Work - or a Contribution incorporated within the Work constitutes direct - or contributory patent infringement, then any patent licenses - granted to You under this License for that Work shall terminate - as of the date such litigation is filed. - - 4. Redistribution. You may reproduce and distribute copies of the - Work or Derivative Works thereof in any medium, with or without - modifications, and in Source or Object form, provided that You - meet the following conditions: - - (a) You must give any other recipients of the Work or - Derivative Works a copy of this License; and - - (b) You must cause any modified files to carry prominent notices - stating that You changed the files; and - - (c) You must retain, in the Source form of any Derivative Works - that You distribute, all copyright, patent, trademark, and - attribution notices from the Source form of the Work, - excluding those notices that do not pertain to any part of - the Derivative Works; and - - (d) If the Work includes a "NOTICE" text file as part of its - distribution, then any Derivative Works that You distribute must - include a readable copy of the attribution notices contained - within such NOTICE file, excluding those notices that do not - pertain to any part of the Derivative Works, in at least one - of the following places: within a NOTICE text file distributed - as part of the Derivative Works; within the Source form or - documentation, if provided along with the Derivative Works; or, - within a display generated by the Derivative Works, if and - wherever such third-party notices normally appear. The contents - of the NOTICE file are for informational purposes only and - do not modify the License. You may add Your own attribution - notices within Derivative Works that You distribute, alongside - or as an addendum to the NOTICE text from the Work, provided - that such additional attribution notices cannot be construed - as modifying the License. - - You may add Your own copyright statement to Your modifications and - may provide additional or different license terms and conditions - for use, reproduction, or distribution of Your modifications, or - for any such Derivative Works as a whole, provided Your use, - reproduction, and distribution of the Work otherwise complies with - the conditions stated in this License. - - 5. Submission of Contributions. Unless You explicitly state otherwise, - any Contribution intentionally submitted for inclusion in the Work - by You to the Licensor shall be under the terms and conditions of - this License, without any additional terms or conditions. - Notwithstanding the above, nothing herein shall supersede or modify - the terms of any separate license agreement you may have executed - with Licensor regarding such Contributions. - - 6. Trademarks. This License does not grant permission to use the trade - names, trademarks, service marks, or product names of the Licensor, - except as required for reasonable and customary use in describing the - origin of the Work and reproducing the content of the NOTICE file. - - 7. Disclaimer of Warranty. Unless required by applicable law or - agreed to in writing, Licensor provides the Work (and each - Contributor provides its Contributions) on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or - implied, including, without limitation, any warranties or conditions - of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A - PARTICULAR PURPOSE. You are solely responsible for determining the - appropriateness of using or redistributing the Work and assume any - risks associated with Your exercise of permissions under this License. - - 8. Limitation of Liability. In no event and under no legal theory, - whether in tort (including negligence), contract, or otherwise, - unless required by applicable law (such as deliberate and grossly - negligent acts) or agreed to in writing, shall any Contributor be - liable to You for damages, including any direct, indirect, special, - incidental, or consequential damages of any character arising as a - result of this License or out of the use or inability to use the - Work (including but not limited to damages for loss of goodwill, - work stoppage, computer failure or malfunction, or any and all - other commercial damages or losses), even if such Contributor - has been advised of the possibility of such damages. - - 9. Accepting Warranty or Additional Liability. While redistributing - the Work or Derivative Works thereof, You may choose to offer, - and charge a fee for, acceptance of support, warranty, indemnity, - or other liability obligations and/or rights consistent with this - License. However, in accepting such obligations, You may act only - on Your own behalf and on Your sole responsibility, not on behalf - of any other Contributor, and only if You agree to indemnify, - defend, and hold each Contributor harmless for any liability - incurred by, or claims asserted against, such Contributor by reason - of your accepting any such warranty or additional liability. - - END OF TERMS AND CONDITIONS - - APPENDIX: How to apply the Apache License to your work. +TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + +1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + +2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + +3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + +4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + +5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + +6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + +7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + +8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + +9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + +END OF TERMS AND CONDITIONS + +APPENDIX: How to apply the Apache License to your work. To apply the Apache License to your work, attach the following boilerplate notice, with the fields enclosed by brackets "[]" @@ -186,16 +186,16 @@ same "printed page" as the copyright notice for easier identification within third-party archives. - Copyright 2026 Quantiles +Copyright 2026 Quantiles, Inc. - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. diff --git a/README.md b/README.md index 661b17f..4c80c8b 100644 --- a/README.md +++ b/README.md @@ -15,7 +15,8 @@ This skill teaches coding agents to do the following: - Use `qt compare ` to analyze changes between evaluation runs. - Use `--json` to inspect structured results. - Use `qt resume` when an evaluation run is interrupted or partially completed. -- Report aggregate metrics, sample-level results, failed samples, regressions, and next steps. +- Use `qt add ` to add a built-in benchmark from the hosted Quantiles registry into a local configuration file. +- Report and analyze aggregate metrics, sample-level results, failed samples, regressions, and next steps. ## Install diff --git a/SKILL.md b/SKILL.md index c88b62c..4ebbe3f 100644 --- a/SKILL.md +++ b/SKILL.md @@ -5,7 +5,7 @@ description: Use when writing, running, inspecting, comparing, resuming, or anal # Quantiles evaluation workflows -Use this skill for Quantiles AI evaluation work. The `qt` CLI is the canonical entrypoint for running benchmarks and evaluations, configuring evaluations and benchmarks, inspecting run results, comparing runs, resuming interrupted evaluations, and executing custom Python evaluation workflows. +Use this skill for Quantiles AI evaluation work. The `qt` CLI is the canonical entrypoint for running benchmarks and evaluations, configuring evaluations and benchmarks, inspecting and analyzing evaluation results, comparing runs, resuming interrupted or failed evaluations, and executing custom Python evaluation workflows. Prefer `qt` CLI commands over manually reading local Quantiles storage files unless the CLI output is insufficient. Never modify local Quantiles storage files, except via `qt` CLI commands. @@ -40,12 +40,12 @@ Follow these rules for all Quantiles work: 1. Use the `qt` CLI as the source of truth. 2. Prefer CLI output over manually reading `.quantiles/` files. Never manually edit or delete `.quantiles/` files unless explicitly asked. -3. Use `--json` for `qt run`, `qt resume`, `qt list`, `qt show`, and `qt compare`. +3. Use `--json` for `qt run`, `qt resume`, `qt list`, `qt show`, `qt compare`, and `qt add`. 4. Report the exact command used. 5. Report the `run_id` after every successful run. -6. Do not claim a demo model run measures real model quality. -7. Do not run evaluations that call external model APIs unless the user explicitly requests one or provides a model name. -8. For external model runs, verify the required provider API key is configured without printing the key value. +6. Do not claim a demo model run measures real AI model quality. +7. Do not run evaluations that call external AI model APIs unless the user explicitly requests one or provides a model name. +8. For external AI model runs, verify the required provider API key is configured without printing the key value. 9. For small sample counts (i.e., when the number of samples in the run is a small fraction of the total samples in the benchmark or evaluation), warn the user to be cautious when interpreting the results given the small sample size. 10. Before saying one run is better than another, check whether the runs are comparable. 11. Never print, log, or expose API key values. @@ -82,7 +82,7 @@ Do not run `qt init` repeatedly without checking whether the repository is alrea See the [model configuration guide](https://quantiles.io/documentation/model-configuration) for supported providers, model identifiers, required credentials, cost and data-handling considerations, and troubleshooting guidance. -Do not call an external model API (e.g., OpenAI or Anthropic) unless the user requests a hosted model run or provides a provider-prefixed model name. Never print API key values. Check whether the required credential is present without revealing it. +Do not call an external AI model API (e.g., OpenAI or Anthropic) unless the user requests a hosted model run or provides a provider-prefixed model name. Never print API key values. Check whether the required credential is present without revealing it. Use the provider prefix in `model` to decide which environment variables to check: @@ -128,70 +128,38 @@ If the required environment variable is missing, stop before calling the externa If the user asks for an evaluation run with a hosted model, but does not specify a sample count, start with a small smoke-test limit unless the user explicitly asks for a larger run. -## Running built-in evaluations +## Running built-in benchmarks -The `qt` CLI includes built-in evaluations such as: +The `qt` CLI includes built-in benchmarks such as: | Eval | Purpose | | ------------------- | --------------------------------------------------------------------------- | | `pubmedqa` | Evaluates model performance on standard healthcare knowledge questions | | `simpleqa-verified` | Evaluates general factual knowledge using an updated SimpleQA-style dataset | -Run a built-in evaluation with: +A list of Quantiles built-in benchmarks is available on the [Quantiles Benchmark Hub](https://quantiles.io/benchmark-hub#built-in). -```bash -qt run --json -``` - -For initial evaluation validation, prefer a small local smoke test: - -```bash -qt run simpleqa-verified --json --input '{"limit":10}' -``` - -If the test succeeds and the user approves a larger evaluation run, remove or increase the `limit`. Before treating a run without `--input` as a full-dataset run, confirm that the evaluation's configuration does not set `samples` or another sample limit: +Run a built-in benchmark with: ```bash -qt run simpleqa-verified --json +qt run --json ``` -The above examples use a demo model, which, as described above, just generates random text. See the "Customizing the model" section below for details on using hosted LLM providers instead. - -### Limiting sample count - -For built-in evaluations, pass a JSON `limit` key through `--input`. In `quantiles.toml`, the equivalent built-in config field is `samples`. - -Example: +Add a built-in benchmark to `quantiles.toml` or `.quantiles.toml` to customize its settings for the current and future runs: ```bash -qt run simpleqa-verified --input '{"limit":100}' --json +qt add --json ``` -Use small limits for smoke tests and larger limits only when the user wants a more complete benchmark run. - -For evaluations that call external model APIs, use a small limit first unless the user explicitly requests a larger run. +Get complete details on [configuring built-in benchmarks](https://quantiles.io/documentation/built-in-benchmarks#apply-persistent-configuration-settings). -### Customizing the model - -If the user asks to use a hosted LLM provider, configure the model either as a persistent default in `quantiles.toml` or `.quantiles.toml` or as a run-specific override through `--input`: - -```bash -qt run --input '{"model":":"}' --json -``` - -See the [configuration documentation](https://quantiles.io/documentation/configuration) for configuring evaluations and the [model configuration guide](https://quantiles.io/documentation/model-configuration) for configuring provider models and credentials. - -Before running an evaluation that calls an external model API, follow the credential checks in the "Secrets and cost guidance" section. For shell command examples, use the project’s documented default model when available; otherwise, use a concrete provider-prefixed model identifier. - -After running, report whether the run used the demo model or a model from an external provider. - -## Custom no-code evaluations +## Custom configuration evaluations Use this section when the user asks to configure a dataset-backed exact-match or multiple-choice evaluation without writing custom evaluation code. -Custom no-code evaluations are configured in `quantiles.toml` or `.quantiles.toml` with `type = "custom_nocode"` and a style table whose type is `"exact_match"` or `"multiple_choice"`. They run inside the `qt` CLI, render each prompt with a Jinja template, call the configured model, and score the response using the configured style. +Custom configuration evaluations are configured in `quantiles.toml` or `.quantiles.toml` with `type = "custom_nocode"` and a style table whose type is `"exact_match"` or `"multiple_choice"`. They run inside the `qt` CLI, render each prompt with a Jinja template, call the configured model, and score the response using the configured style. -Custom no-code evaluations currently load Hugging Face datasets. The `dataset` table requires `name` and optionally accepts `config_name`, `split`, and `revision`. +Custom configuration evaluations currently load Hugging Face datasets. The `dataset` table requires `name` and optionally accepts `config_name`, `split`, and `revision`. For exact-match scoring, configure the expected-answer field inside `style`: @@ -242,7 +210,7 @@ A multiple-choice template receives normalized `choices` in addition to `row`: Choices can come from one array or label-keyed object field through `style.choices.column`, or from multiple scalar fields through `style.choices.columns`. Array-backed rows may contain fewer choices than the configured labels, but not more. The answer source must use exactly one of `label_column`, `index_column`, or `correct_choice_column`; `index_column` uses a zero-based index unless `style.answer.index_base` is configured. Choice labels must be nonempty and unique; when using `style.choices.columns`, the columns and labels must have equal lengths. `correct_choice_column` requires `style.choices.columns` and must name one of those columns. To shuffle choices deterministically, configure `style.shuffle.seed_column` with a stable scalar field from each row. -Run the benchmark with: +Run the evaluation with: ```bash qt run --json @@ -254,7 +222,7 @@ Then inspect the run: qt show --json ``` -When configuring or reviewing a custom no-code evaluation: +When configuring or reviewing a custom configuration evaluation: - Confirm `prompt_template_file` exists and contains valid Jinja before running. - Confirm the template's `row` fields and all style-specific fields exist in the dataset. @@ -267,7 +235,7 @@ When configuring or reviewing a custom no-code evaluation: - Report the run ID, model, dataset, prompt template path, sample count, accuracy, correctness counts, parse-rate metrics, and latency metrics when summarizing results. - For multiple-choice evaluations, also report the relevant macro, weighted, per-label, support, and confusion-matrix metrics. -For more information, see the [custom no-code documentation](https://quantiles.io/documentation/custom-evaluations/custom-nocode-evaluations). +For more information, see the [custom configuration documentation](https://quantiles.io/documentation/custom-evaluations/custom-nocode-evaluations). ## Custom code evaluations @@ -346,6 +314,55 @@ When converting an ad-hoc evaluation script into a Quantiles evaluation: First preserve behavior, then make it durable and observable. +## Customize evaluation settings + +First, identify the evaluation type and whether the change should apply to one evaluation run or future runs: + +- Use `--input` for one-time overrides. +- Use `quantiles.toml` or `.quantiles.toml` for persistent settings. + +Supported fields and override behavior vary by evaluation type. + +### Limit the sample count + +For a one-time limit, built-in benchmarks and custom configuration evaluations accept `limit` through `--input`: + +```bash +qt run --input '{"limit":10}' --json +``` + +For persistent limits: + +- Built-in benchmarks use `samples`. +- Custom configuration evaluations use `limit`. +- Custom-code evaluations define their own input fields. Pass `limit` only if the evaluation code supports it. + +### Change the AI model + +Before calling an external model API, follow the credential and cost checks in “Secrets and cost guidance.” + +Built-in benchmarks and custom configuration evaluations accept a provider-prefixed `model` in one of two ways: + +- Add a one-time `--input` override: + + ```bash + qt run --input '{"model":":"}' --json + ``` + + Override behavior differs by evaluation type: + - For built-in benchmarks, `--input` replaces the configured run settings. Include all required overrides in the same JSON object. + - For custom configuration evaluations, supported `--input` fields override only the specified values for the current run. Other settings continue to come from the configuration file. + - For custom-code evaluations, configure the model in the evaluation code. Use `[benchmarks..input]` or `--input` only if the evaluation explicitly supports a model field. + +- Add the model to the evaluation’s configuration: + + ```toml + [benchmarks.] + model = ":" + ``` + + See the [model configuration guide](https://quantiles.io/documentation/model-configuration) for provider and credential setup and the [configuration documentation](https://quantiles.io/documentation/configuration) for evaluation settings. + ## JSON output Always pass `--json` to the following commands: @@ -354,7 +371,8 @@ Always pass `--json` to the following commands: - `qt resume --json` - `qt list --json` - `qt show --json` -- `qt compare --json` +- `qt compare --json` +- `qt add --json` `qt run` returns structured data that includes the `run_id` and evaluation name. Inspect the output to extract the requested information and perform the requested analysis. @@ -432,7 +450,7 @@ If there are many runs, narrow by benchmark name, evaluation name, timestamp, st The `qt` CLI can compare two evaluation runs. To compare runs, first identify the two `run_id` values, then run the following: ```bash -qt compare --json +qt compare --json ``` Before comparing runs, confirm that the comparison is apples-to-apples using the following criteria: @@ -443,7 +461,7 @@ Before comparing runs, confirm that the comparison is apples-to-apples using the - Same split or sample selection, if available - Same scorer or grader - Same metric definitions -- Same model-vs-demo sampler setup +- Same hosted AI model or demo model (`model = "random"`) - Same input JSON except for the intended changed variable - Same provider settings, temperature, seed, or decoding parameters, if available @@ -464,6 +482,8 @@ When comparing, report the following information, if available, along with any a Use `qt resume` to recover the same evaluation run after an operational failure such as a timeout, rate limit, network error, local process exit, or interruption. Resume only when continuing the same evaluation configuration. Start a new run when intentionally changing the model, prompt, dataset, rubric, workflow input, step input, or another behavior-changing value. Do not resume a completed run. +For built-in benchmarks loaded from the Quantiles registry originally, resuming requires internet access to retrieve the benchmark configuration. + Inspect the run and resolve the failure before resuming it: ```bash @@ -488,7 +508,7 @@ When summarizing a completed evaluation run, include both the raw result and a h For every completed run summary, include: - The evaluation name and, when inferable, what capability or behavior it is intended to measure. -- The model name and whether the run used a model from an external provider or `demo-builtin`. +- The model name and whether the run used a model from an external provider or the demo model (`model="random"`). - The primary metrics and their values, using the metric names emitted by the evaluation. - The sample count and whether the run appears to be a smoke test, partial run, or full configured run. - A plain-English interpretation of what the result suggests. @@ -498,7 +518,7 @@ For every completed run summary, include: Do not impose a generic meaning on metrics. First use the evaluation's own documentation, config, emitted metric names, and sample-level outputs to infer what the metrics mean. If metric semantics are unclear, say so and give a cautious interpretation rather than pretending the meaning is known. -Never claim `demo-builtin` results measure real model quality. For demo runs, interpret only benchmark execution, metric shape, sample distribution, and storage behavior. +Fully analyze demo-model runs, but clearly label them as workflow-validation runs. Do not interpret their results as evidence of hosted AI model quality or representative benchmark performance. ## Suggested reporting template @@ -551,6 +571,10 @@ Keep the response concise unless the user asks for detailed sample-level analysi Use these checks when something fails. +### Identify the error + +Use `qt list` to find the evaluation run ID when needed. Inspect the run with `qt show --json`, identify any reported errors, and resolve them. + ### `qt` command not found Check installation and PATH: @@ -572,10 +596,14 @@ Then verify: command -v qt && qt --help ``` -### Built-in benchmark ran but results look random +### Evaluation ran but results look random Check whether the run used the demo model. Demo model output is expected to be random or fake and should only be used for overall validation and smoke testing. +### Built-in benchmark run fails + +First, check whether the built-in benchmark is defined in a local `quantiles.toml` or `.quantiles.toml` file. If found, validate its configuration. Otherwise, confirm that the benchmark exists in the hosted Quantiles benchmark registry and that internet access is available to retrieve its configuration. Continue troubleshooting based on the result. + ### External model evaluation fails immediately Check the provider credential described in the "Secrets and cost guidance" section without revealing its value. Then verify that the `model` value uses a supported provider prefix and an exact model identifier available to the provider account.