merge v0.4.2

requirements-cpu.txt

@@ -2,5 +2,5 @@
 -r requirements-common.txt
 
 # Dependencies for x86_64 CPUs
-torch == 2.2.1+cpu
+torch == 2.3.0+cpu
 triton >= 2.2.0  # FIXME(woosuk): This is a hack to avoid import error.
\ No newline at end of file

requirements-cuda.txt

@@ -5,5 +5,5 @@
 ray >= 2.9
 nvidia-ml-py # for pynvml package
 vllm-nccl-cu12>=2.18,<2.19  # for downloading nccl library
-torch == 2.2.1
-xformers == 0.0.25  # Requires PyTorch 2.2.1
+torch == 2.3.0
+xformers == 0.0.26.post1  # Requires PyTorch 2.3.0

requirements-dev.txt

@@ -14,19 +14,17 @@ types-setuptools
 
 # testing
 pytest
-tensorizer==2.9.0a0
+tensorizer==2.9.0
 pytest-forked
 pytest-asyncio
 pytest-rerunfailures
 pytest-shard
 httpx
 einops # required for MPT
-openai
 requests
 ray
 peft
 awscli
-ai2-olmo # required for OLMo
 
 # Benchmarking
 aiohttp

tests/async_engine/test_async_llm_engine.py

@@ -91,4 +91,6 @@ async def test_new_requests_event():
     assert engine.engine.step_calls == old_step_calls + 1
 
     engine = MockAsyncLLMEngine(worker_use_ray=True, engine_use_ray=True)
+    assert engine.get_model_config() is not None
     assert engine.get_tokenizer() is not None
+    assert engine.get_decoding_config() is not None

tests/async_engine/test_merge_async_iterators.py

+import asyncio
+from typing import AsyncIterator, Tuple
+
+import pytest
+
+from vllm.utils import merge_async_iterators
+
+
+@pytest.mark.asyncio
+async def test_merge_async_iterators():
+
+    async def mock_async_iterator(idx: int) -> AsyncIterator[str]:
+        try:
+            while True:
+                yield f"item from iterator {idx}"
+                await asyncio.sleep(0.1)
+        except asyncio.CancelledError:
+            pass
+
+    iterators = [mock_async_iterator(i) for i in range(3)]
+    merged_iterator: AsyncIterator[Tuple[int, str]] = merge_async_iterators(
+        *iterators)
+
+    async def stream_output(generator: AsyncIterator[Tuple[int, str]]):
+        async for idx, output in generator:
+            print(f"idx: {idx}, output: {output}")
+
+    task = asyncio.create_task(stream_output(merged_iterator))
+    await asyncio.sleep(0.5)
+    task.cancel()
+    with pytest.raises(asyncio.CancelledError):
+        await task
+
+    for iterator in iterators:
+        try:
+            await asyncio.wait_for(anext(iterator), 1)
+        except StopAsyncIteration:
+            # All iterators should be cancelled and print this message.
+            print("Iterator was cancelled normally")
+        except (Exception, asyncio.CancelledError) as e:
+            raise AssertionError() from e