From 185ff00c62853c376d1a7dbeac98d0e60517b2fc Mon Sep 17 00:00:00 2001
From: alkinun <alkinunl@gmail.com>
Date: Thu, 28 May 2026 12:28:52 +0300
Subject: [PATCH] Fix non-streaming GGUF chat completion usage (#5781)

* Fix GGUF non-stream chat completion usage

* Handle nullable GGUF completion usage

---------

Co-authored-by: Lee Jackson <130007945+Imagineer99@users.noreply.github.com>
Co-authored-by: Roland Tannous <115670425+rolandtannous@users.noreply.github.com>
---
 studio/backend/routes/inference.py            | 14 +++-
 .../tests/test_gguf_completion_usage.py       | 78 +++++++++++++++++++
 2 files changed, 91 insertions(+), 1 deletion(-)
 create mode 100644 studio/backend/tests/test_gguf_completion_usage.py

diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index 259337616c..7d1c7b2488 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -3091,9 +3091,12 @@ async def openai_chat_completions(
         else:
             try:
                 full_text = ""
+                completion_usage = None
                 for token in gguf_generate():
                     if isinstance(token, dict):
-                        continue  # skip metadata dict in non-streaming path
+                        if token.get("type") == "metadata":
+                            completion_usage = token.get("usage")
+                        continue
                     full_text = token
 
                 response = ChatCompletion(
@@ -3106,6 +3109,15 @@ async def openai_chat_completions(
                             finish_reason = "stop",
                         )
                     ],
+                    usage = CompletionUsage(
+                        prompt_tokens = (completion_usage or {}).get("prompt_tokens")
+                        or 0,
+                        completion_tokens = (completion_usage or {}).get(
+                            "completion_tokens"
+                        )
+                        or 0,
+                        total_tokens = (completion_usage or {}).get("total_tokens") or 0,
+                    ),
                 )
                 return JSONResponse(content = response.model_dump())
 
diff --git a/studio/backend/tests/test_gguf_completion_usage.py b/studio/backend/tests/test_gguf_completion_usage.py
new file mode 100644
index 0000000000..b8cfaee7c7
--- /dev/null
+++ b/studio/backend/tests/test_gguf_completion_usage.py
@@ -0,0 +1,78 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved.
+
+"""Regression tests for GGUF non-streaming chat completion usage."""
+
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+from auth.authentication import get_current_subject
+import routes.inference as inference_route
+
+
+class _GgufBackend:
+    is_loaded = True
+    model_identifier = "test/model.gguf"
+    _is_audio = False
+    is_vision = False
+    supports_tools = False
+
+    def __init__(self, usage):
+        self.usage = usage
+
+    def generate_chat_completion(self, **kwargs):
+        yield "answer"
+        yield {
+            "type": "metadata",
+            "usage": self.usage,
+            "timings": {"prompt_n": 23, "predicted_n": 1283},
+        }
+
+
+def _request_completion(monkeypatch, usage):
+    monkeypatch.setattr(
+        inference_route, "get_llama_cpp_backend", lambda: _GgufBackend(usage)
+    )
+    monkeypatch.setattr(
+        inference_route, "_effective_enable_tools", lambda payload: False
+    )
+
+    app = FastAPI()
+    app.include_router(inference_route.router)
+    app.dependency_overrides[get_current_subject] = lambda: "test-user"
+
+    return TestClient(app).post(
+        "/chat/completions",
+        json = {
+            "messages": [{"role": "user", "content": "Why is the sky blue?"}],
+            "stream": False,
+        },
+    )
+
+
+def test_non_streaming_gguf_completion_includes_generated_usage(monkeypatch):
+    response = _request_completion(
+        monkeypatch,
+        {"prompt_tokens": 23, "completion_tokens": 1283, "total_tokens": 1306},
+    )
+
+    assert response.status_code == 200
+    assert response.json()["usage"] == {
+        "prompt_tokens": 23,
+        "completion_tokens": 1283,
+        "total_tokens": 1306,
+    }
+
+
+def test_non_streaming_gguf_completion_defaults_nullable_usage_to_zero(monkeypatch):
+    response = _request_completion(
+        monkeypatch,
+        {"prompt_tokens": None, "completion_tokens": 1283, "total_tokens": None},
+    )
+
+    assert response.status_code == 200
+    assert response.json()["usage"] == {
+        "prompt_tokens": 0,
+        "completion_tokens": 1283,
+        "total_tokens": 0,
+    }