[Frontend] Add sampling params to v1/audio/transcriptions endpoint (#16591)

Signed-off-by: Jannis Schönleber <joennlae@gmail.com>
Signed-off-by: NickLucche <nlucches@redhat.com>
Co-authored-by: Jannis Schönleber <joennlae@gmail.com>

tests/entrypoints/openai/test_transcription_validation.py

@@ -192,3 +192,36 @@ async def test_stream_options(winning_call):
                 else:
                     continuous = continuous and hasattr(chunk, 'usage')
             assert final and continuous
+
+
+@pytest.mark.asyncio
+async def test_sampling_params(mary_had_lamb):
+    """
+    Compare sampling with params and greedy sampling to assert results
+    are different when extreme sampling parameters values are picked. 
+    """
+    model_name = "openai/whisper-small"
+    server_args = ["--enforce-eager"]
+    with RemoteOpenAIServer(model_name, server_args) as remote_server:
+        client = remote_server.get_async_client()
+        transcription = await client.audio.transcriptions.create(
+            model=model_name,
+            file=mary_had_lamb,
+            language="en",
+            temperature=0.8,
+            extra_body=dict(seed=42,
+                            repetition_penalty=1.9,
+                            top_k=12,
+                            top_p=0.4,
+                            min_p=0.5,
+                            frequency_penalty=1.8,
+                            presence_penalty=2.0))
+
+        greedy_transcription = await client.audio.transcriptions.create(
+            model=model_name,
+            file=mary_had_lamb,
+            language="en",
+            temperature=0.0,
+            extra_body=dict(seed=42))
+
+        assert greedy_transcription.text != transcription.text