* Only write streaming provider-proxy cache body on finishWe were previously writing the current body when the stream wasdropped, even if it never finished (e.g. due to a timeout).As a result, we could end up writing an invalid body to diskdue to an *earlier* request from the e2e tests (if the e2e testshit an internal timeout and dropped the stream), overwritingthe good cache line from a later successful test run* Wait for provider-proxy write before sending back responseThis should help prevent race conditions involving client-side retries* Fix typoCo-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>* Fix clippy* Forward Cloudflare R2 env vars into fixtures containerSince these were unset inside the container, we were fallingback to the slow dev-endpoint download, rather than the fastaws cli download* Install aws cli in fixtures Dockerfile* added multi-threaded runtime for test that was exploding---------Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>Co-authored-by: Viraj Mehta <viraj@tensorzero.com>

…es (#5189)* added a get config route* added migration for tags and missing file* removed failing assertion* added json schema derives to uninitialized variant and evaluation types* made the migration public from tensorzero-core crate

* Stream inferences and datapoints in UI* Stream inferences and datapoints in UI* Stream inferences and datapoints in UI* Stream inferences and datapoints in UI* Stream inferences and datapoints in UI---------Co-authored-by: Viraj Mehta <viraj@tensorzero.com>

This should reduce the flakiness of this test (we were previouslyonly discarding the thought blocks in some cases)

* added a get config route* added migration for tags and missing file* removed failing assertion* made the migration public from tensorzero-core crate* added generic action handler* added e2e tests* added rust client method* use rust client in tests* fixed PR comments

Closes#4783.Co-authored-by: Gabriel Bianconi <1275491+GabrielBianconi@users.noreply.github.com>

* Add support for extra_headers/extra_body in relay modeThe relay gateway now forwards these options to the downstreamgateway (after performing variant-level filtering on the relaygateway).* Fix typoCo-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>* Remove collectCo-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>* Run fmt* Add some unit tests* Add more unit tests---------Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Co-authored-by: Aaron Hill <aaron@tensorzero.com>

…modal (#5197)* Add searchable evaluation and variant selectors to Launch Evaluation modal- Add generic combobox UI components (Combobox, ComboboxInput, ComboboxContent, useCombobox hook)- Add EvaluationSelector component that wraps the generic Combobox- Add VariantSelector component that wraps the generic Combobox- Update LaunchEvaluationModal to use searchable selectors instead of static Select- Update Input component focus style to use border darkening instead of ring* Add documentation comment to useCombobox hook* Update e2e tests for Combobox selectors in Launch Evaluation modal

)* e2e tests: feedback: construct payloads using structs + serializeContributes to#4710.* rely on Serialize---------Co-authored-by: Aaron Hill <aaron@tensorzero.com>

* Prevent duplicate fake-path templates from being loadedWhen loading templates from the config, any fake-path keysshould be unique (these potentially come from agent-generatedvariant configs, which will have inline templates)* Fix typoCo-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>* Fix test---------Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* Clean up Docker Compose for UI* Lower log level for config message

* Tag filter for != should still require the tag to be set* Add test

…5201)Contributes to#4710.

…5219)

* Initial partial implementation of betting-based confidence sequences* Fix bets to use previous time step's variance estimate* Add updates to wealth processes* Add last step of confidence sequence calculation from hedged wealth process* Change product to log-sum-exp for numerical stability* Add computation of point estimator for the mean* Combine weighting and hedged wealth process computation* Move bet calculation to helper function, add bet truncation* Tweak comment string for clarity* Add module-level and type-level docstrings* Remove max vs combo hedging choice, explain use of max in docstring* For computing mean estimate, restrict search to inside confidence interval for efficiency* Add unit tests, fix search for interval endpoints* Tweak test and test comments* Fix regularized mean value, add regression tests with known values* Move test, clarify comment strings in another test* Add test with known confidence sequence values* Change tests to use non-constant observations for variance fluctuations* Rename module to contrast with asymptotic_confidence_sequences.rs* Add input validation and associated tests* Remove unnecessary iterator method for m-values* Make return type a Result, use anyhow for errors* Create enums and structs* Add check_topk_stopping(), allow dead code as needed* Add epsilon argument, changed from allow to expect dead_code* Add wip note to docstring* Add tests using non-zero epsilon tolerance* Add more tests with k_min not equal to k_max* Add docstrings to tests* Change return value when k_min > num_variants* Tweak docstrings* Remove enum that will be included in a separate PR* Add VariantStatus enum* Refactor evaluations to accept a vector of variants for batch processing* Remove changes from topk.rs for now* Use partition_point() instead of binary search for finding confidence sequence bounds* Make enum for specifying grid of points where wealth processes are calculated* Move argument validation for WealthProcessGridPoints to new constructor methods* Fix bug where variant could count itself when checking how many variants it beats* Add variant_names option to CLI and python client* Fix tests to use new WealthProcessGridPoints enum* Don't multiply num_datapoints by num_variants* Update some tests to use variant_names instead of variant_name* Add deprecation warning for variant_name argument* Add deprecation warning for variant_name arg to python client* Factor out batch evals functionality into helper function* Revert change to allow passing multiple variants to run_evaluations()* Add variant back into tracing span* Pre-resolve datapoint inputs to minimize clones

* Migrate getWorkflowEvaluationProjects* Add clickhouse e2e tests

)* added the key info to the request extensions from tensorzero-auth* put in a single request extension

…5216)Bumps the rust-dependencies group with 4 updates in the / directory: [reqwest](https://github.com/seanmonstar/reqwest), [minijinja](https://github.com/mitsuhiko/minijinja), [rcgen](https://github.com/rustls/rcgen) and [tree-sitter](https://github.com/tree-sitter/tree-sitter).Updates `reqwest` from 0.12.25 to 0.12.26- [Release notes](https://github.com/seanmonstar/reqwest/releases)- [Changelog](https://github.com/seanmonstar/reqwest/blob/master/CHANGELOG.md)- [Commits](seanmonstar/reqwest@v0.12.25...v0.12.26)Updates `minijinja` from 2.13.0 to 2.14.0- [Release notes](https://github.com/mitsuhiko/minijinja/releases)- [Changelog](https://github.com/mitsuhiko/minijinja/blob/main/CHANGELOG.md)- [Commits](mitsuhiko/minijinja@2.13.0...2.14.0)Updates `rcgen` from 0.14.5 to 0.14.6- [Release notes](https://github.com/rustls/rcgen/releases)- [Commits](rustls/rcgen@v0.14.5...v0.14.6)Updates `tree-sitter` from 0.25.10 to 0.26.3- [Release notes](https://github.com/tree-sitter/tree-sitter/releases)- [Commits](tree-sitter/tree-sitter@v0.25.10...v0.26.3)---updated-dependencies:- dependency-name: reqwest  dependency-version: 0.12.26  dependency-type: direct:production  update-type: version-update:semver-patch  dependency-group: rust-dependencies- dependency-name: minijinja  dependency-version: 2.14.0  dependency-type: direct:production  update-type: version-update:semver-minor  dependency-group: rust-dependencies- dependency-name: rcgen  dependency-version: 0.14.6  dependency-type: direct:production  update-type: version-update:semver-patch  dependency-group: rust-dependencies- dependency-name: tree-sitter  dependency-version: 0.26.3  dependency-type: direct:production  update-type: version-update:semver-minor  dependency-group: rust-dependencies...Signed-off-by: dependabot[bot] <support@github.com>Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* Claude 3.7 -> 4.5* Claude 3.7 -> 4.5* Claude 3.7 -> 4.5* Claude 3.7 -> 4.5

* Tag filter for != should still require the tag to be set* Add test* Fix other NULL filters* Fix* Fix

* Migrate countInferencesForEpisode* Update route

This setting has significant performance implications,and is useful to know when viewing the soon-to-be-added'overhead' metric

Frontend

Contributes to#3460.

This test is constantly flaking