Merge commit 'ab86335760ebb441574eb47f886fa1ee302e2131' into concedo_experimental

# Conflicts:
#	.github/workflows/release.yml
#	examples/retrieval/retrieval.cpp
#	examples/simple-chat/simple-chat.cpp
#	ggml/src/ggml-opencl/ggml-opencl.cpp
#	ggml/src/ggml-sycl/ggml-sycl.cpp
#	requirements/requirements-convert_hf_to_gguf.txt
#	requirements/requirements-convert_hf_to_gguf_update.txt
#	requirements/requirements-convert_lora_to_gguf.txt
#	tools/run/run.cpp

tools/server/tests/unit/test_chat_completion.py

@@ -71,8 +71,14 @@ def test_chat_completion_stream(system_prompt, user_prompt, max_tokens, re_conte
     })
     content = ""
     last_cmpl_id = None
-    for data in res:
+    for i, data in enumerate(res):
         choice = data["choices"][0]
+        if i == 0:
+            # Check first role message for stream=True
+            assert choice["delta"]["content"] == ""
+            assert choice["delta"]["role"] == "assistant"
+        else:
+            assert "role" not in choice["delta"]
         assert data["system_fingerprint"].startswith("b")
         assert "gpt-3.5" in data["model"] # DEFAULT_OAICOMPAT_MODEL, maybe changed in the future
         if last_cmpl_id is None:
@@ -242,12 +248,18 @@ def test_chat_completion_with_timings_per_token():
         "stream": True,
         "timings_per_token": True,
     })
-    for data in res:
-        assert "timings" in data
-        assert "prompt_per_second" in data["timings"]
-        assert "predicted_per_second" in data["timings"]
-        assert "predicted_n" in data["timings"]
-        assert data["timings"]["predicted_n"] <= 10
+    for i, data in enumerate(res):
+        if i == 0:
+            # Check first role message for stream=True
+            assert data["choices"][0]["delta"]["content"] == ""
+            assert data["choices"][0]["delta"]["role"] == "assistant"
+        else:
+            assert "role" not in data["choices"][0]["delta"]
+            assert "timings" in data
+            assert "prompt_per_second" in data["timings"]
+            assert "predicted_per_second" in data["timings"]
+            assert "predicted_n" in data["timings"]
+            assert data["timings"]["predicted_n"] <= 10
 
 
 def test_logprobs():
@@ -295,17 +307,23 @@ def test_logprobs_stream():
     )
     output_text = ''
     aggregated_text = ''
-    for data in res:
+    for i, data in enumerate(res):
         choice = data.choices[0]
-        if choice.finish_reason is None:
-            if choice.delta.content:
-                output_text += choice.delta.content
-            assert choice.logprobs is not None
-            assert choice.logprobs.content is not None
-            for token in choice.logprobs.content:
-                aggregated_text += token.token
-                assert token.logprob <= 0.0
-                assert token.bytes is not None
-                assert token.top_logprobs is not None
-                assert len(token.top_logprobs) > 0
+        if i == 0:
+            # Check first role message for stream=True
+            assert choice.delta.content == ""
+            assert choice.delta.role == "assistant"
+        else:
+            assert choice.delta.role is None
+            if choice.finish_reason is None:
+                if choice.delta.content:
+                    output_text += choice.delta.content
+                assert choice.logprobs is not None
+                assert choice.logprobs.content is not None
+                for token in choice.logprobs.content:
+                    aggregated_text += token.token
+                    assert token.logprob <= 0.0
+                    assert token.bytes is not None
+                    assert token.top_logprobs is not None
+                    assert len(token.top_logprobs) > 0
     assert aggregated_text == output_text