Clean up nld flags by evelyn-with-warp · Pull Request #10671 · warpdotdev/warp

evelyn-with-warp · 2026-05-11T21:02:49Z

Remove the NLD-related feature flags and fasttext classifier
Now that the ONNX/BertTiny classifier is the production default:

Description

Linked Issue

The linked issue is labeled ready-to-spec or ready-to-implement.
Where appropriate, screenshots or a short video of the implementation are included below (especially for user-visible or UI changes).

Testing

https://www.loom.com/share/2d0bc3c4834e43928356266039da1b40

I have manually tested my changes locally with ./script/run

Screenshots / Videos

Agent Mode

Warp Agent Mode - This PR was created via Warp's AI Agent Mode

Reviewers

@vorporeal @szgupta

Remove the NLD-related feature flags now that the ONNX/BertTiny classifier is the production default: - Drop FeatureFlag::NLDClassifierModelEnabled and FeatureFlag::NldImprovements (including the DOGFOOD_FLAGS entry and the cfg-gated wiring in app/src/lib.rs). - Drop the nld_fasttext_model, nld_onnx_model, and nld_improvements cargo features. Enable input_classifier/onnx_candle directly on the app dependency so the BertTiny model still compiles in. - Simplify InputClassifierModel::new to always try OnnxClassifier::new(BertTiny) with HeuristicClassifier as the only fallback; remove the fasttext branch and the is_nld_classifier_enabled helper that read the EnableNLDClassifierModel user preference. - Inline the new two-option UDI segmented-control rendering in universal_developer_input.rs; the legacy build_renderable_option_config body and the duplicate build_new_renderable_option_config function are gone. - InputBufferSubmitted telemetry now fires unconditionally; the corresponding EnablementState flips to Always. - Collapse the input_mode_prefix_tests macro from 8 cases (nld_improvements x udi x InputType) to 4 (udi x InputType). - Drop the stale TODO comment in input_model.rs that referenced NldImprovements. - Remove ,nld_improvements from the macOS, Linux, and Windows bundle scripts on every channel/artifact branch. Co-Authored-By: Oz <oz-agent@warp.dev>

oz-for-oss · 2026-05-11T21:03:03Z

@evelyn-with-warp

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

oz-for-oss

Overview

This PR removes the NLD-related feature flags and cargo feature plumbing, makes the ONNX/BertTiny classifier the app default with the heuristic classifier as fallback, updates UDI rendering to the promoted two-option control, and sends InputBufferSubmitted telemetry unconditionally.

Concerns

No blocking correctness or security concerns found in the annotated diff.

Verdict

Found: 0 critical, 0 important, 0 suggestions

Approve

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

The fasttext classifier was never enabled in any shipping bundle and is fully superseded by the ONNX BertTiny classifier hard-coded in `InputClassifierModel`. Drop the module, its embedded model, the `fasttext` cargo feature and its optional deps, and the fasttext branches from the `evaluate` binary. Co-Authored-By: Oz <oz-agent@warp.dev>

szgupta · 2026-05-11T21:49:04Z


                    // If we have history entries (i.e., a live session), check for
                    // close matches to short-circuit as shell input.
-                    // TODO(vorporeal): decide if we still want to do this with NldImprovements.


Are we tracking this somewhere? Linear?

This should be related from internal report: https://linear.app/warpdotdev/issue/QUALITY-577/improve-nld-on-dependency-for-shell-command-history
side note: looks like we are doing it today with nld improvement, but we might hope to do a better job in short-circuit with user command history cc @vorporeal

szgupta · 2026-05-11T21:49:50Z

        {
            // If we're submitting an AI query, we want to send telemetry for the input type.
-            if FeatureFlag::NldImprovements.is_enabled() {
+            {


nit: remove unnecessary scoping

szgupta · 2026-05-11T21:52:04Z

 if [[ "$ARTIFACT" == "cli" ]]; then
  FEATURES="$FEATURES,standalone"
 elif [[ "$ARTIFACT" == "app" ]]; then
-  FEATURES="$FEATURES,gui,nld_improvements"


Per https://warpdev.slack.com/archives/C08QY6ZBK2A/p1778531062702359?thread_ts=1778529278.316949&cid=C08QY6ZBK2A, how are we ensuring that CLI builds don't bundle the model?

CLI build will bundle the model, as a price to pay for a cleaner state (with as minimum feature flags as possible);
Here's the change 592c7b6 I was thinking of for long term: gate all app ml models together and let CLI builds ignore them; The benefit is we could edit the ml_model gate freely without changing bundle scripts, and meanwhile protected cli/oz builds from them

it looks like we are replacing nld_improvment flag with bundled_all_ml_models flag, hopefully this will have long term benefit for readability and maintenance.

CLI build will bundle the model, as a price to pay for a cleaner state

i don't think we're ok with paying this price. 100% of the time we would take slightly more complex code in the bundling side of things over bloating the release size by 17MB with no functional benefit.

@vorporeal i agree for this. So current solution is we use bundled_app_ml_model flag, and the thought here is going forward this flag could handle all ml models (not only for nld work) that cli/oz should be excluded from

evelyn-with-warp · 2026-05-11T23:36:05Z

 lsp-types = "0.97.0"
 indexmap = { version = "2.0.2", features = ["serde"] }
-input_classifier = { workspace = true, features = ["onnx_candle"] }
+input_classifier.workspace = true


features will be captured in ml_model_input_classifier_onnx

…ml_models Co-Authored-By: Oz <oz-agent@warp.dev>

evelyn-with-warp · 2026-05-12T17:32:25Z

+    "nld_onnx_classifier",
+]
+nld_classifier_v2 = [
+    "nld_classifier_v1",


cc @szgupta this will be modified in the following PR when we add the new classifier

szgupta · 2026-05-12T18:13:00Z

    fn tokenizer_path(&self) -> &'static str {
        match self {
-            Model::BertTiny => "bert_tiny_tokenizer.json",
+            Model::BertTinyV1 => "bert_tiny_tokenizer.json",


will both v1 and v2 use the same tokenizer?

@vorporeal

Remove the NLD-related feature flags and fasttext classifier Now that the ONNX/BertTiny classifier is the production default: ## Description  ## Linked Issue  - [ ] The linked issue is labeled `ready-to-spec` or `ready-to-implement`. - [ ] Where appropriate, screenshots or a short video of the implementation are included below (especially for user-visible or UI changes). ## Testing  https://www.loom.com/share/2d0bc3c4834e43928356266039da1b40 - [x] I have manually tested my changes locally with `./script/run` ### Screenshots / Videos  ## Agent Mode - [x] Warp Agent Mode - This PR was created via Warp's AI Agent Mode  ## Reviewers @vorporeal @szgupta --------- Co-authored-by: Oz <oz-agent@warp.dev>

## Description Remove the stale `nld_improvements` Cargo feature from the Nix package build. `nld_improvements` was removed in #10671, but the flake still passed it through `buildFeatures`. That causes `nix run .#warp-terminal-experimental` to fail during Cargo feature validation. Closes #10979. ## Linked Issue  - [ ] The linked issue is labeled `ready-to-spec` or `ready-to-implement`. - [ ] Where appropriate, screenshots or a short video of the implementation are included below (especially for user-visible or UI changes). ## Testing  - [ ] I have manually tested my changes locally with `./script/run` Validated locally with `nix run .#warp-terminal-experimental` ### Screenshots / Videos  ## Agent Mode - [ ] Warp Agent Mode - This PR was created via Warp's AI Agent Mode

@vorporeal

Remove the NLD-related feature flags and fasttext classifier Now that the ONNX/BertTiny classifier is the production default: ## Description  ## Linked Issue  - [ ] The linked issue is labeled `ready-to-spec` or `ready-to-implement`. - [ ] Where appropriate, screenshots or a short video of the implementation are included below (especially for user-visible or UI changes). ## Testing  https://www.loom.com/share/2d0bc3c4834e43928356266039da1b40 - [x] I have manually tested my changes locally with `./script/run` ### Screenshots / Videos  ## Agent Mode - [x] Warp Agent Mode - This PR was created via Warp's AI Agent Mode  ## Reviewers @vorporeal @szgupta --------- Co-authored-by: Oz <oz-agent@warp.dev>

cla-bot Bot added the cla-signed label May 11, 2026

oz-for-oss Bot reviewed May 11, 2026

View reviewed changes

szgupta reviewed May 11, 2026

View reviewed changes

evelyn-with-warp commented May 11, 2026

View reviewed changes

refactor ml model feature flags by gating all of them behind bundled_…

5fc917a

…ml_models Co-Authored-By: Oz <oz-agent@warp.dev>

evelyn-with-warp force-pushed the evelyn/cleanup-nld-flags branch from 592c7b6 to 5fc917a Compare May 11, 2026 23:36

clean up

1057664

evelyn-with-warp requested a review from szgupta May 11, 2026 23:46

rename feature name to be nld driven

1207c88

evelyn-with-warp commented May 12, 2026

View reviewed changes

Rename input classifier for version consistency

077f357

szgupta approved these changes May 12, 2026

View reviewed changes

evelyn-with-warp enabled auto-merge (squash) May 12, 2026 18:14

evelyn-with-warp merged commit 2c1f204 into master May 12, 2026
25 of 26 checks passed

evelyn-with-warp deleted the evelyn/cleanup-nld-flags branch May 12, 2026 18:16

bet4it mentioned this pull request Jun 2, 2026

Fix Nix build features #12070

Merged

4 tasks

Conversation

evelyn-with-warp commented May 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Linked Issue

Testing

Screenshots / Videos

Agent Mode

Reviewers

Uh oh!

oz-for-oss Bot commented May 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

oz-for-oss Bot left a comment

Choose a reason for hiding this comment

Overview

Concerns

Verdict

Uh oh!

Choose a reason for hiding this comment

Uh oh!

evelyn-with-warp May 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

evelyn-with-warp May 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

evelyn-with-warp May 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

evelyn-with-warp commented May 11, 2026 •

edited

Loading

oz-for-oss Bot commented May 11, 2026 •

edited

Loading

evelyn-with-warp May 11, 2026 •

edited

Loading

evelyn-with-warp May 11, 2026 •

edited

Loading

evelyn-with-warp May 12, 2026 •

edited

Loading