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/basic_correctness/test_chunked_prefill.py

@@ -55,7 +55,6 @@ def test_models(
     )
     vllm_outputs = vllm_model.generate_greedy(example_prompts, max_tokens)
     del vllm_model
-    print(vllm_outputs[0])
 
     for i in range(len(example_prompts)):
         hf_output_ids, hf_output_str = hf_outputs[i]

tests/conftest.py

@@ -296,6 +296,7 @@ class VllmRunner:
         tensor_parallel_size: int = 1,
         block_size: int = 16,
         enable_chunked_prefill: bool = False,
+        swap_space=4,
         **kwargs,
     ) -> None:
         self.model = LLM(
@@ -303,7 +304,7 @@ class VllmRunner:
             tokenizer=tokenizer_name,
             trust_remote_code=True,
             dtype=dtype,
-            swap_space=0,
+            swap_space=swap_space,
             disable_log_stats=disable_log_stats,
             tensor_parallel_size=tensor_parallel_size,
             max_model_len=max_model_len,