Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	.github/labeler.yml
#	.github/workflows/release.yml
#	examples/debug/debug.cpp
#	ggml/src/ggml-cuda/common.cuh
#	ggml/src/ggml-cuda/mmq.cuh
#	ggml/src/ggml-webgpu/ggml-webgpu.cpp
#	src/llama-vocab.cpp
#	tests/test-backend-ops.cpp
#	tests/test-chat.cpp
#	tests/test-json-schema-to-grammar.cpp
#	tools/mtmd/CMakeLists.txt

tools/server/tests/unit/test_completion.py

@@ -135,7 +135,7 @@ def test_completion_stream_with_openai_library_stops():
     client = OpenAI(api_key="dummy", base_url=f"http://{server.server_host}:{server.server_port}/v1")
     res = client.completions.create(
         model="davinci-002",
-        prompt="System: You are helpfull assistant.\nAssistant:\nHey! How could I help?\nUser:\nTell me a joke.\nAssistant:\n",
+        prompt="System: You are helpful assistant.\nAssistant:\nHey! How could I help?\nUser:\nTell me a joke.\nAssistant:\n",
         stop=["User:\n", "Assistant:\n"],
         max_tokens=200,
         stream=True,

tools/server/tests/unit/test_ignore_eos.py

@@ -0,0 +1,43 @@
+import pytest
+from utils import *
+
+server = ServerPreset.tinyllama2()
+
+
+@pytest.fixture(autouse=True)
+def create_server():
+    global server
+    server = ServerPreset.tinyllama2()
+
+
+def test_ignore_eos_populates_logit_bias():
+    """ignore_eos=true must add EOG logit biases to generation_settings."""
+    global server
+    server.start()
+    res = server.make_request("POST", "/completion", data={
+        "n_predict": 8,
+        "prompt": "Once upon a time",
+        "ignore_eos": True,
+        "temperature": 0.0,
+    })
+    assert res.status_code == 200
+    # EOG token biases must be present with -inf bias
+    logit_bias = res.body["generation_settings"]["logit_bias"]
+    assert len(logit_bias) > 0
+    for entry in logit_bias:
+        assert entry["bias"] is None  # null in JSON represents -inf
+
+
+def test_ignore_eos_false_no_logit_bias():
+    """ignore_eos=false (default) must NOT add EOG logit biases."""
+    global server
+    server.start()
+    res = server.make_request("POST", "/completion", data={
+        "n_predict": 8,
+        "prompt": "Once upon a time",
+        "ignore_eos": False,
+        "temperature": 0.0,
+    })
+    assert res.status_code == 200
+    logit_bias = res.body["generation_settings"]["logit_bias"]
+    assert len(logit_bias) == 0