From d8a4627355eb8b108e842534296b48c57c59a837 Mon Sep 17 00:00:00 2001
From: Daniel Han <danielhanchen@gmail.com>
Date: Sun, 24 May 2026 13:44:38 +0000
Subject: [PATCH] Studio: widen scale type, preserve significant ws in stops,
 Kimi parity

Round 2 reviewer feedback:

- studio/frontend/src/features/chat/types/api.ts: `OpenAIChatCompletionsRequest.service_tier` did not include `"scale"`, so the request builder in chat-adapter.ts failed typecheck after the runtime ServiceTier union widened (`Type 'ServiceTier | undefined' is not assignable...`). Widen the type to match the SDK and keep the typecheck green.

- studio/frontend/src/components/ui/stop-sequences-input.tsx: the chip editor used `draft.trim()` for storage, which silently mutated semantically meaningful stops like " End", "### ", and "\n\n". Keep the whitespace-only rejection (Anthropic 400s on those, OpenAI silently drops them) but persist the raw draft so leading/trailing whitespace inside otherwise-meaningful stops survives.

- studio/backend/core/inference/external_provider.py: the Kimi web-search bypass dropped a single string `stop="\n\n"` via `stop.strip()` while the normal default OAI-compat path forwards it verbatim. Mirror the default path's behavior here so kimi-with-search and kimi-without-search apply the same rules (asymmetric provider-path fix flagged in round-2 review).
---
 studio/backend/core/inference/external_provider.py | 10 ++++++++--
 .../src/components/ui/stop-sequences-input.tsx     | 14 ++++++++++----
 studio/frontend/src/features/chat/types/api.ts     | 12 ++++++++++--
 3 files changed, 28 insertions(+), 8 deletions(-)

diff --git a/studio/backend/core/inference/external_provider.py b/studio/backend/core/inference/external_provider.py
index 002e2548c6..d55f132487 100644
--- a/studio/backend/core/inference/external_provider.py
+++ b/studio/backend/core/inference/external_provider.py
@@ -912,9 +912,15 @@ class ExternalProviderClient:
         if seed is not None:
             body["seed"] = seed
         if stop:
+            # Mirror the default OAI-compat path's stop handling exactly
+            # so Kimi-with-search and Kimi-without-search apply the
+            # same rules — a single string is forwarded verbatim and
+            # lists are deduped + truncated to OpenAI's 4-entry cap.
+            # Earlier the bypass dropped whitespace-only strings here
+            # while the normal path forwarded them, which was an
+            # asymmetric provider-path fix.
             if isinstance(stop, str):
-                if stop.strip():
-                    body["stop"] = stop
+                body["stop"] = stop
             elif isinstance(stop, list):
                 sequences = list(
                     dict.fromkeys(s for s in stop if isinstance(s, str) and s)
diff --git a/studio/frontend/src/components/ui/stop-sequences-input.tsx b/studio/frontend/src/components/ui/stop-sequences-input.tsx
index e3c2d8a561..5d65fdd36a 100644
--- a/studio/frontend/src/components/ui/stop-sequences-input.tsx
+++ b/studio/frontend/src/components/ui/stop-sequences-input.tsx
@@ -38,14 +38,20 @@ export function StopSequencesInput({
   const atCap = value.length >= maxEntries;
 
   function commitDraft() {
-    const trimmed = draft.trim();
-    if (!trimmed) return;
+    // Reject chips that are empty or contain ONLY whitespace
+    // (Anthropic 400s on those and OpenAI silently drops them), but
+    // preserve significant leading/trailing whitespace inside otherwise
+    // -meaningful stops like " END", "### ", or "\n\n" — stop matching
+    // is exact, so stripping would silently change the semantics. The
+    // backend re-validates per-provider before the request hits the
+    // wire.
+    if (!draft || !draft.trim()) return;
     if (atCap) return;
-    if (value.includes(trimmed)) {
+    if (value.includes(draft)) {
       setDraft("");
       return;
     }
-    onChange([...value, trimmed]);
+    onChange([...value, draft]);
     setDraft("");
   }
 
diff --git a/studio/frontend/src/features/chat/types/api.ts b/studio/frontend/src/features/chat/types/api.ts
index e94f9dca7c..7c5b5bf99f 100644
--- a/studio/frontend/src/features/chat/types/api.ts
+++ b/studio/frontend/src/features/chat/types/api.ts
@@ -280,14 +280,22 @@ export interface OpenAIChatCompletionsRequest {
   stop?: string[];
   /**
    * Provider service tier. Anthropic accepts `auto|standard_only`;
-   * OpenAI Chat accepts `auto|default|flex|priority|scale`; OpenAI
-   * Responses accepts `auto|default|flex|priority`.
+   * OpenAI Chat + Responses both accept
+   * `auto|default|flex|scale|priority` per the live `openai-python`
+   * SDK (`src/openai/types/responses/response_create_params.py`
+   * declares `Optional[Literal["auto", "default", "flex", "scale",
+   * "priority"]]`). The wire-side helper in
+   * `studio/backend/core/inference/external_provider.py` drops values
+   * that a given provider does not accept; this union stays permissive
+   * so the request-builder typechecks against
+   * `InferenceParams.serviceTier` without per-provider narrowing.
    */
   service_tier?:
     | "auto"
     | "default"
     | "flex"
     | "priority"
+    | "scale"
     | "standard_only";
   /**
    * Whether the provider may dispatch tool calls in parallel.