Revert "rename PromptInputs and inputs with backward compatibility (vllm-project#8760) (vllm-project#8810)

Author: simon-mo

benchmarks/benchmark_latency.py

@@ -11,7 +11,7 @@
 
 from vllm import LLM, SamplingParams
 from vllm.engine.arg_utils import DEVICE_OPTIONS, EngineArgs
-from vllm.inputs import PromptType
+from vllm.inputs import PromptInputs
 from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
 from vllm.utils import FlexibleArgumentParser
 
@@ -61,7 +61,7 @@ def main(args: argparse.Namespace):
     dummy_prompt_token_ids = np.random.randint(10000,
                                                size=(args.batch_size,
                                                      args.input_len))
-    dummy_prompts: List[PromptType] = [{
+    dummy_inputs: List[PromptInputs] = [{
         "prompt_token_ids": batch
     } for batch in dummy_prompt_token_ids.tolist()]
 
@@ -74,13 +74,13 @@ def run_to_completion(profile_dir: Optional[str] = None):
                     ],
                     on_trace_ready=torch.profiler.tensorboard_trace_handler(
                         str(profile_dir))) as p:
-                llm.generate(dummy_prompts,
+                llm.generate(dummy_inputs,
                              sampling_params=sampling_params,
                              use_tqdm=False)
             print(p.key_averages())
         else:
             start_time = time.perf_counter()
-            llm.generate(dummy_prompts,
+            llm.generate(dummy_inputs,
                          sampling_params=sampling_params,
                          use_tqdm=False)
             end_time = time.perf_counter()

docs/source/dev/multimodal/multimodal_index.rst

@@ -8,7 +8,7 @@ Multi-Modality
 vLLM provides experimental support for multi-modal models through the :mod:`vllm.multimodal` package.
 
 Multi-modal inputs can be passed alongside text and token prompts to :ref:`supported models <supported_vlms>`
-via the ``multi_modal_data`` field in :class:`vllm.inputs.PromptType`.
+via the ``multi_modal_data`` field in :class:`vllm.inputs.PromptInputs`.
 
 Currently, vLLM only has built-in support for image data. You can extend vLLM to process additional modalities
 by following :ref:`this guide <adding_multimodal_plugin>`.

docs/source/dev/offline_inference/llm_inputs.rst

@@ -1,7 +1,7 @@
 LLM Inputs
 ==========
 
-.. autodata:: vllm.inputs.PromptType
+.. autodata:: vllm.inputs.PromptInputs
 
 .. autoclass:: vllm.inputs.TextPrompt
     :show-inheritance:

docs/source/models/vlm.rst

@@ -27,7 +27,7 @@ The :class:`~vllm.LLM` class can be instantiated in much the same way as languag
     We have removed all vision language related CLI args in the ``0.5.1`` release. **This is a breaking change**, so please update your code to follow
     the above snippet. Specifically, ``image_feature_size`` can no longer be specified as we now calculate that internally for each model.
 
-To pass an image to the model, note the following in :class:`vllm.inputs.PromptType`:
+To pass an image to the model, note the following in :class:`vllm.inputs.PromptInputs`:
 
 * ``prompt``: The prompt should follow the format that is documented on HuggingFace.
 * ``multi_modal_data``: This is a dictionary that follows the schema defined in :class:`vllm.multimodal.MultiModalDataDict`. 

tests/async_engine/test_async_llm_engine.py

@@ -86,19 +86,17 @@ class MockAsyncLLMEngine(AsyncLLMEngine):
 
 @pytest.mark.asyncio
 async def test_new_requests_event():
-    params = SamplingParams()
-
     engine = MockAsyncLLMEngine()
     engine.start_background_loop()
     await asyncio.sleep(0.01)
     assert engine.engine.step_calls == 0
 
-    await engine.add_request("1", "", params)
+    await engine.add_request("1", "", None)
     await asyncio.sleep(0.01)
     assert engine.engine.add_request_calls == 1
     assert engine.engine.step_calls == 1
 
-    await engine.add_request("2", "", params)
+    await engine.add_request("2", "", None)
     engine.engine.generate("2")
     await asyncio.sleep(0)
     await asyncio.sleep(0)
@@ -113,7 +111,7 @@ async def test_new_requests_event():
     await asyncio.sleep(0.001)
     assert engine.engine.step_calls == old_step_calls
 
-    await engine.add_request("3", "", params)
+    await engine.add_request("3", "", None)
     await asyncio.sleep(0.01)
     assert engine.engine.add_request_calls == 3
     assert engine.engine.step_calls == old_step_calls + 1

tests/entrypoints/llm/test_encode.py

@@ -49,6 +49,21 @@ def assert_outputs_equal(o1: List[EmbeddingRequestOutput],
     assert [o.outputs for o in o1] == [o.outputs for o in o2]
 
 
+@pytest.mark.skip_global_cleanup
+@pytest.mark.parametrize('prompt', PROMPTS)
+def test_v1_v2_api_consistency_single_prompt_string(llm: LLM, prompt):
+    pooling_params = PoolingParams()
+
+    with pytest.warns(DeprecationWarning, match="'prompts'"):
+        v1_output = llm.encode(prompts=prompt, pooling_params=pooling_params)
+
+    v2_output = llm.encode(prompt, pooling_params=pooling_params)
+    assert_outputs_equal(v1_output, v2_output)
+
+    v2_output = llm.encode({"prompt": prompt}, pooling_params=pooling_params)
+    assert_outputs_equal(v1_output, v2_output)
+
+
 @pytest.mark.skip_global_cleanup
 @pytest.mark.parametrize('prompt_token_ids', TOKEN_IDS)
 def test_v1_v2_api_consistency_single_prompt_tokens(llm: LLM,
@@ -64,6 +79,25 @@ def test_v1_v2_api_consistency_single_prompt_tokens(llm: LLM,
     assert_outputs_equal(v1_output, v2_output)
 
 
+@pytest.mark.skip_global_cleanup
+def test_v1_v2_api_consistency_multi_prompt_string(llm: LLM):
+    pooling_params = PoolingParams()
+
+    with pytest.warns(DeprecationWarning, match="'prompts'"):
+        v1_output = llm.encode(prompts=PROMPTS, pooling_params=pooling_params)
+
+    v2_output = llm.encode(PROMPTS, pooling_params=pooling_params)
+    assert_outputs_equal(v1_output, v2_output)
+
+    v2_output = llm.encode(
+        [{
+            "prompt": p
+        } for p in PROMPTS],
+        pooling_params=pooling_params,
+    )
+    assert_outputs_equal(v1_output, v2_output)
+
+
 @pytest.mark.skip_global_cleanup
 def test_v1_v2_api_consistency_multi_prompt_tokens(llm: LLM):
     pooling_params = PoolingParams()

tests/entrypoints/llm/test_generate.py

@@ -47,6 +47,23 @@ def assert_outputs_equal(o1: List[RequestOutput], o2: List[RequestOutput]):
     assert [o.outputs for o in o1] == [o.outputs for o in o2]
 
 
+@pytest.mark.skip_global_cleanup
+@pytest.mark.parametrize('prompt', PROMPTS)
+def test_v1_v2_api_consistency_single_prompt_string(llm: LLM, prompt):
+    sampling_params = SamplingParams(temperature=0.0, top_p=1.0)
+
+    with pytest.warns(DeprecationWarning, match="'prompts'"):
+        v1_output = llm.generate(prompts=prompt,
+                                 sampling_params=sampling_params)
+
+    v2_output = llm.generate(prompt, sampling_params=sampling_params)
+    assert_outputs_equal(v1_output, v2_output)
+
+    v2_output = llm.generate({"prompt": prompt},
+                             sampling_params=sampling_params)
+    assert_outputs_equal(v1_output, v2_output)
+
+
 @pytest.mark.skip_global_cleanup
 @pytest.mark.parametrize('prompt_token_ids', TOKEN_IDS)
 def test_v1_v2_api_consistency_single_prompt_tokens(llm: LLM,
@@ -62,6 +79,26 @@ def test_v1_v2_api_consistency_single_prompt_tokens(llm: LLM,
     assert_outputs_equal(v1_output, v2_output)
 
 
+@pytest.mark.skip_global_cleanup
+def test_v1_v2_api_consistency_multi_prompt_string(llm: LLM):
+    sampling_params = SamplingParams(temperature=0.0, top_p=1.0)
+
+    with pytest.warns(DeprecationWarning, match="'prompts'"):
+        v1_output = llm.generate(prompts=PROMPTS,
+                                 sampling_params=sampling_params)
+
+    v2_output = llm.generate(PROMPTS, sampling_params=sampling_params)
+    assert_outputs_equal(v1_output, v2_output)
+
+    v2_output = llm.generate(
+        [{
+            "prompt": p
+        } for p in PROMPTS],
+        sampling_params=sampling_params,
+    )
+    assert_outputs_equal(v1_output, v2_output)
+
+
 @pytest.mark.skip_global_cleanup
 def test_v1_v2_api_consistency_multi_prompt_tokens(llm: LLM):
     sampling_params = SamplingParams(temperature=0.0, top_p=1.0)

tests/mq_llm_engine/test_error_handling.py

@@ -61,15 +61,15 @@ async def test_evil_forward(tmp_socket):
 
         # Throws an error in first forward pass.
         with pytest.raises(RAISED_ERROR):
-            async for _ in client.generate(prompt="Hello my name is",
+            async for _ in client.generate(inputs="Hello my name is",
                                            sampling_params=SamplingParams(),
                                            request_id=uuid.uuid4()):
                 pass
         assert client.errored
 
         # Engine is errored, should get ENGINE_DEAD_ERROR.
         with pytest.raises(MQEngineDeadError):
-            async for _ in client.generate(prompt="Hello my name is",
+            async for _ in client.generate(inputs="Hello my name is",
                                            sampling_params=SamplingParams(),
                                            request_id=uuid.uuid4()):
                 pass
@@ -118,7 +118,7 @@ async def test_failed_health_check(tmp_socket):
 
         # Generate call should throw ENGINE_DEAD_ERROR
         with pytest.raises(MQEngineDeadError):
-            async for _ in client.generate(prompt="Hello my name is",
+            async for _ in client.generate(inputs="Hello my name is",
                                            sampling_params=SamplingParams(),
                                            request_id=uuid.uuid4()):
                 pass
@@ -160,7 +160,7 @@ async def test_failed_abort(tmp_socket):
         # with reference to the original KeyError("foo")
         with pytest.raises(MQEngineDeadError) as execinfo:
             async for _ in client.generate(
-                    prompt="Hello my name is",
+                    inputs="Hello my name is",
                     sampling_params=SamplingParams(max_tokens=10),
                     request_id=uuid.uuid4()):
                 pass
@@ -183,7 +183,7 @@ async def test_bad_request(tmp_socket):
 
         # Invalid request should fail, but not crash the server.
         with pytest.raises(ValueError):
-            async for _ in client.generate(prompt="Hello my name is",
+            async for _ in client.generate(inputs="Hello my name is",
                                            sampling_params=SamplingParams(),
                                            request_id="abcd-1",
                                            lora_request=LoRARequest(
@@ -192,7 +192,7 @@ async def test_bad_request(tmp_socket):
                 pass
 
         # This request should be okay.
-        async for _ in client.generate(prompt="Hello my name is",
+        async for _ in client.generate(inputs="Hello my name is",
                                        sampling_params=SamplingParams(),
                                        request_id="abcd-2"):
             pass

tests/mq_llm_engine/utils.py

@@ -20,7 +20,7 @@ async def generate(
     count = 0
     async for out in client.generate(
             request_id=request_id,
-            prompt="Hello my name is Robert and",
+            inputs="Hello my name is Robert and",
             sampling_params=SamplingParams(max_tokens=num_tokens,
                                            temperature=0)):
 

vllm/__init__.py

@@ -5,7 +5,7 @@
 from vllm.engine.llm_engine import LLMEngine
 from vllm.entrypoints.llm import LLM
 from vllm.executor.ray_utils import initialize_ray_cluster
-from vllm.inputs import PromptType, TextPrompt, TokensPrompt
+from vllm.inputs import PromptInputs, TextPrompt, TokensPrompt
 from vllm.model_executor.models import ModelRegistry
 from vllm.outputs import (CompletionOutput, EmbeddingOutput,
                           EmbeddingRequestOutput, RequestOutput)
@@ -19,7 +19,7 @@
     "__version_tuple__",
     "LLM",
     "ModelRegistry",
-    "PromptType",
+    "PromptInputs",
     "TextPrompt",
     "TokensPrompt",
     "SamplingParams",