From f4466bae65f0890e7556102b7729e3aaff1c15bf Mon Sep 17 00:00:00 2001
From: Alexander Matveev <59768536+alexm-neuralmagic@users.noreply.github.com>
Date: Tue, 3 Dec 2024 05:33:10 -0500
Subject: [PATCH] [V1] VLM - preprocessor hashing

Signed-off-by: Roger Wang <ywang@roblox.com>
Co-authored-by: Michael Goin <michael@neuralmagic.com>
Co-authored-by: Roger Wang <ywang@roblox.com>
Signed-off-by: Alexander Matveev <alexm@neuralmagic.com>
---
 examples/offline_inference_vision_language.py | 126 ++++++++++++--
 vllm/config.py                                |  10 +-
 vllm/engine/arg_utils.py                      |   8 +
 vllm/v1/engine/__init__.py                    |   3 +-
 vllm/v1/engine/core.py                        |  13 +-
 vllm/v1/engine/mm_input_mapper.py             | 158 ++++++++++++++++--
 vllm/v1/engine/processor.py                   |  35 ++--
 vllm/v1/utils.py                              |  21 +++
 8 files changed, 326 insertions(+), 48 deletions(-)

diff --git a/examples/offline_inference_vision_language.py b/examples/offline_inference_vision_language.py
index c6a274ee5894b..5e210126dc8fe 100644
--- a/examples/offline_inference_vision_language.py
+++ b/examples/offline_inference_vision_language.py
@@ -5,6 +5,8 @@
 For most models, the prompt format should follow corresponding examples
 on HuggingFace model repository.
 """
+import random
+
 from transformers import AutoTokenizer
 
 from vllm import LLM, SamplingParams
@@ -23,7 +25,9 @@ def run_llava(question: str, modality: str):
 
     prompt = f"USER: <image>\n{question}\nASSISTANT:"
 
-    llm = LLM(model="llava-hf/llava-1.5-7b-hf", max_model_len=4096)
+    llm = LLM(model="llava-hf/llava-1.5-7b-hf",
+              max_model_len=4096,
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
     stop_token_ids = None
     return llm, prompt, stop_token_ids
 
@@ -33,7 +37,9 @@ def run_llava_next(question: str, modality: str):
     assert modality == "image"
 
     prompt = f"[INST] <image>\n{question} [/INST]"
-    llm = LLM(model="llava-hf/llava-v1.6-mistral-7b-hf", max_model_len=8192)
+    llm = LLM(model="llava-hf/llava-v1.6-mistral-7b-hf",
+              max_model_len=8192,
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
     stop_token_ids = None
     return llm, prompt, stop_token_ids
 
@@ -44,7 +50,9 @@ def run_llava_next_video(question: str, modality: str):
     assert modality == "video"
 
     prompt = f"USER: <video>\n{question} ASSISTANT:"
-    llm = LLM(model="llava-hf/LLaVA-NeXT-Video-7B-hf", max_model_len=8192)
+    llm = LLM(model="llava-hf/LLaVA-NeXT-Video-7B-hf",
+              max_model_len=8192,
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
     stop_token_ids = None
     return llm, prompt, stop_token_ids
 
@@ -61,7 +69,8 @@ def run_llava_onevision(question: str, modality: str):
         <|im_start|>assistant\n"
 
     llm = LLM(model="llava-hf/llava-onevision-qwen2-7b-ov-hf",
-              max_model_len=16384)
+              max_model_len=16384,
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
     stop_token_ids = None
     return llm, prompt, stop_token_ids
 
@@ -71,7 +80,10 @@ def run_fuyu(question: str, modality: str):
     assert modality == "image"
 
     prompt = f"{question}\n"
-    llm = LLM(model="adept/fuyu-8b", max_model_len=2048, max_num_seqs=2)
+    llm = LLM(model="adept/fuyu-8b",
+              max_model_len=2048,
+              max_num_seqs=2,
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
     stop_token_ids = None
     return llm, prompt, stop_token_ids
 
@@ -107,6 +119,7 @@ def run_phi3v(question: str, modality: str):
         max_num_seqs=2,
         # Note - mm_processor_kwargs can also be passed to generate/chat calls
         mm_processor_kwargs={"num_crops": 16},
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
     stop_token_ids = None
     return llm, prompt, stop_token_ids
@@ -118,7 +131,8 @@ def run_paligemma(question: str, modality: str):
 
     # PaliGemma has special prompt format for VQA
     prompt = "caption en"
-    llm = LLM(model="google/paligemma-3b-mix-224")
+    llm = LLM(model="google/paligemma-3b-mix-224",
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
     stop_token_ids = None
     return llm, prompt, stop_token_ids
 
@@ -128,7 +142,9 @@ def run_chameleon(question: str, modality: str):
     assert modality == "image"
 
     prompt = f"{question}<image>"
-    llm = LLM(model="facebook/chameleon-7b", max_model_len=4096)
+    llm = LLM(model="facebook/chameleon-7b",
+              max_model_len=4096,
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
     stop_token_ids = None
     return llm, prompt, stop_token_ids
 
@@ -154,6 +170,7 @@ def run_minicpmv(question: str, modality: str):
         max_model_len=4096,
         max_num_seqs=2,
         trust_remote_code=True,
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
     # NOTE The stop_token_ids are different for various versions of MiniCPM-V
     # 2.0
@@ -186,6 +203,7 @@ def run_h2ovl(question: str, modality: str):
         model=model_name,
         trust_remote_code=True,
         max_model_len=8192,
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
 
     tokenizer = AutoTokenizer.from_pretrained(model_name,
@@ -211,6 +229,7 @@ def run_internvl(question: str, modality: str):
         model=model_name,
         trust_remote_code=True,
         max_model_len=4096,
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
 
     tokenizer = AutoTokenizer.from_pretrained(model_name,
@@ -241,6 +260,7 @@ def run_nvlm_d(question: str, modality: str):
         trust_remote_code=True,
         max_model_len=4096,
         tensor_parallel_size=4,
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
 
     tokenizer = AutoTokenizer.from_pretrained(model_name,
@@ -260,7 +280,8 @@ def run_blip2(question: str, modality: str):
     # BLIP-2 prompt format is inaccurate on HuggingFace model repository.
     # See https://huggingface.co/Salesforce/blip2-opt-2.7b/discussions/15#64ff02f3f8cf9e4f5b038262 #noqa
     prompt = f"Question: {question} Answer:"
-    llm = LLM(model="Salesforce/blip2-opt-2.7b")
+    llm = LLM(model="Salesforce/blip2-opt-2.7b",
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
     stop_token_ids = None
     return llm, prompt, stop_token_ids
 
@@ -274,6 +295,7 @@ def run_qwen_vl(question: str, modality: str):
         trust_remote_code=True,
         max_model_len=1024,
         max_num_seqs=2,
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
 
     prompt = f"{question}Picture 1: <img></img>\n"
@@ -296,6 +318,7 @@ def run_qwen2_vl(question: str, modality: str):
             "min_pixels": 28 * 28,
             "max_pixels": 1280 * 28 * 28,
         },
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
 
     prompt = ("<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n"
@@ -315,6 +338,7 @@ def run_pixtral_hf(question: str, modality: str):
     llm = LLM(
         model=model_name,
         max_model_len=8192,
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
 
     prompt = f"<s>[INST]{question}\n[IMG][/INST]"
@@ -338,6 +362,7 @@ def run_mllama(question: str, modality: str):
         max_model_len=4096,
         max_num_seqs=16,
         enforce_eager=True,
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
 
     prompt = f"<|image|><|begin_of_text|>{question}"
@@ -355,6 +380,7 @@ def run_molmo(question, modality):
         model=model_name,
         trust_remote_code=True,
         dtype="bfloat16",
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
 
     prompt = question
@@ -371,7 +397,8 @@ def run_glm4v(question: str, modality: str):
               max_model_len=2048,
               max_num_seqs=2,
               trust_remote_code=True,
-              enforce_eager=True)
+              enforce_eager=True,
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
     prompt = question
     stop_token_ids = [151329, 151336, 151338]
     return llm, prompt, stop_token_ids
@@ -394,6 +421,7 @@ def run_idefics3(question: str, modality: str):
                 "longest_edge": 3 * 364
             },
         },
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
     prompt = (
         f"<|begin_of_text|>User:<image>{question}<end_of_utterance>\nAssistant:"
@@ -410,7 +438,8 @@ def run_aria(question: str, modality: str):
     llm = LLM(model=model_name,
               tokenizer_mode="slow",
               trust_remote_code=True,
-              dtype="bfloat16")
+              dtype="bfloat16",
+              mm_cache_preprocessor=args.mm_cache_preprocessor)
 
     prompt = (f"<|im_start|>user\n<fim_prefix><|img|><fim_suffix>\n{question}"
               "<|im_end|>\n<|im_start|>assistant\n")
@@ -430,6 +459,7 @@ def run_mantis(question: str, modality: str):
         model="TIGER-Lab/Mantis-8B-siglip-llama3",
         max_model_len=4096,
         hf_overrides={"architectures": ["MantisForConditionalGeneration"]},
+        mm_cache_preprocessor=args.mm_cache_preprocessor,
     )
     stop_token_ids = [128009]
     return llm, prompt, stop_token_ids
@@ -494,6 +524,35 @@ def get_multi_modal_input(args):
     raise ValueError(msg)
 
 
+def apply_image_repeat(image_repeat_prob, num_prompts, data, prompt, modality):
+    """Repeats images with provided probability of "image_repeat_prob". 
+    Used to simulate hit/miss for the MM preprocessor cache.
+    """
+    assert (image_repeat_prob <= 1.0 and image_repeat_prob >= 0)
+    no_yes = [0, 1]
+    probs = [1.0 - image_repeat_prob, image_repeat_prob]
+
+    inputs = []
+    cur_image = data
+    for i in range(num_prompts):
+        if image_repeat_prob is not None:
+            res = random.choices(no_yes, probs)[0]
+            if res == 0:
+                # No repeat => Modify one pixel
+                cur_image = cur_image.copy()
+                new_val = (i // 256 // 256, i // 256, i % 256)
+                cur_image.putpixel((0, 0), new_val)
+
+        inputs.append({
+            "prompt": prompt,
+            "multi_modal_data": {
+                modality: cur_image
+            }
+        })
+
+    return inputs
+
+
 def main(args):
     model = args.model_type
     if model not in model_example_map:
@@ -524,14 +583,29 @@ def main(args):
 
     else:
         # Batch inference
-        inputs = [{
-            "prompt": prompt,
-            "multi_modal_data": {
-                modality: data
-            },
-        } for _ in range(args.num_prompts)]
+        if args.image_repeat_prob is not None:
+            # Repeat images with specified probability of "image_repeat_prob"
+            inputs = apply_image_repeat(args.image_repeat_prob,
+                                        args.num_prompts, data, prompt,
+                                        modality)
+        else:
+            # Use the same image for all prompts
+            inputs = [{
+                "prompt": prompt,
+                "multi_modal_data": {
+                    modality: data
+                },
+            } for _ in range(args.num_prompts)]
+
+    if args.time_generate:
+        import time
+        start_time = time.time()
+        outputs = llm.generate(inputs, sampling_params=sampling_params)
+        elapsed_time = time.time() - start_time
+        print("-- generate time = {}".format(elapsed_time))
 
-    outputs = llm.generate(inputs, sampling_params=sampling_params)
+    else:
+        outputs = llm.generate(inputs, sampling_params=sampling_params)
 
     for o in outputs:
         generated_text = o.outputs[0].text
@@ -561,5 +635,23 @@ def main(args):
                         type=int,
                         default=16,
                         help='Number of frames to extract from the video.')
+
+    parser.add_argument(
+        '--image-repeat-prob',
+        type=float,
+        default=None,
+        help='Simulates the hit-ratio for multi-modal preprocessor cache'
+        ' (if enabled)')
+
+    parser.add_argument(
+        '--mm-cache-preprocessor',
+        action='store_true',
+        help='If True, enable caching of multi-modal preprocessor/mapper.')
+
+    parser.add_argument(
+        '--time-generate',
+        action='store_true',
+        help='If True, then print the total generate() call time')
+
     args = parser.parse_args()
     main(args)
diff --git a/vllm/config.py b/vllm/config.py
index 5fb9563fcf3a3..4b8a639ce2fe0 100644
--- a/vllm/config.py
+++ b/vllm/config.py
@@ -133,6 +133,9 @@ class ModelConfig:
             HuggingFace config.
         mm_processor_kwargs: Arguments to be forwarded to the model's processor
             for multi-modal data, e.g., image processor.
+        mm_cache_preprocessor: If true, then enables caching of the multi-modal 
+            preprocessor/mapper. Otherwise, the mapper executes each time, and 
+            for better performance consider enabling frontend process.
         override_neuron_config: Initialize non default neuron config or
             override default neuron config that are specific to Neuron devices,
             this argument will be used to configure the neuron config that
@@ -171,6 +174,7 @@ def __init__(
             config_format: ConfigFormat = ConfigFormat.AUTO,
             hf_overrides: Optional[HfOverrides] = None,
             mm_processor_kwargs: Optional[Dict[str, Any]] = None,
+            mm_cache_preprocessor: bool = False,
             override_neuron_config: Optional[Dict[str, Any]] = None,
             override_pooler_config: Optional["PoolerConfig"] = None) -> None:
         self.model = model
@@ -237,6 +241,7 @@ def __init__(
         self.dtype = _get_and_verify_dtype(self.hf_text_config, dtype)
         self.use_async_output_proc = use_async_output_proc
         self.mm_processor_kwargs = mm_processor_kwargs
+        self.mm_cache_preprocessor = mm_cache_preprocessor
 
         # Set enforce_eager to False if the value is unset.
         if self.enforce_eager is None:
@@ -2600,9 +2605,10 @@ def __str__(self):
             f"enable_prefix_caching={self.cache_config.enable_prefix_caching}, "
             f"chunked_prefill_enabled={self.scheduler_config.chunked_prefill_enabled}, "  # noqa
             f"use_async_output_proc={self.model_config.use_async_output_proc}, "
+            f"mm_cache_preprocessor={self.model_config.mm_cache_preprocessor!r}, "  # noqa
             f"mm_processor_kwargs={self.model_config.mm_processor_kwargs}, "
-            f"pooler_config={self.model_config.pooler_config!r},"
-            f" compilation_config={self.compilation_config!r}")
+            f"pooler_config={self.model_config.pooler_config!r}, "
+            f"compilation_config={self.compilation_config!r}")
 
 
 _current_vllm_config: Optional[VllmConfig] = None
diff --git a/vllm/engine/arg_utils.py b/vllm/engine/arg_utils.py
index 3db069ec64ee4..06a5b544fd57c 100644
--- a/vllm/engine/arg_utils.py
+++ b/vllm/engine/arg_utils.py
@@ -143,6 +143,7 @@ class EngineArgs:
     tokenizer_pool_extra_config: Optional[Dict[str, Any]] = None
     limit_mm_per_prompt: Optional[Mapping[str, int]] = None
     mm_processor_kwargs: Optional[Dict[str, Any]] = None
+    mm_cache_preprocessor: bool = False
     enable_lora: bool = False
     enable_lora_bias: bool = False
     max_loras: int = 1
@@ -590,6 +591,12 @@ def add_cli_args(parser: FlexibleArgumentParser) -> FlexibleArgumentParser:
             type=json.loads,
             help=('Overrides for the multimodal input mapping/processing, '
                   'e.g., image processor. For example: {"num_crops": 4}.'))
+        parser.add_argument(
+            '--mm-cache-preprocessor',
+            action='store_true',
+            help='If true, then enables caching of the multi-modal '
+            'preprocessor/mapper. Otherwise, the mapper executes each time'
+            ', and for better performance consider enabling frontend process.')
 
         # LoRA related configs
         parser.add_argument('--enable-lora',
@@ -962,6 +969,7 @@ def create_model_config(self) -> ModelConfig:
             use_async_output_proc=not self.disable_async_output_proc,
             config_format=self.config_format,
             mm_processor_kwargs=self.mm_processor_kwargs,
+            mm_cache_preprocessor=self.mm_cache_preprocessor,
             override_neuron_config=self.override_neuron_config,
             override_pooler_config=self.override_pooler_config,
         )
diff --git a/vllm/v1/engine/__init__.py b/vllm/v1/engine/__init__.py
index 3cf0e610ae7af..abeea052c1fa5 100644
--- a/vllm/v1/engine/__init__.py
+++ b/vllm/v1/engine/__init__.py
@@ -35,7 +35,8 @@ class EngineCoreRequest:
     # always be tokenized?
     prompt: Optional[str]
     prompt_token_ids: List[int]
-    mm_inputs: Optional[List[MultiModalKwargs]]
+    mm_inputs: Optional[List[Optional[MultiModalKwargs]]]
+    mm_hashes: Optional[List[Optional[str]]]
     mm_placeholders: Optional[MultiModalPlaceholderDict]
     sampling_params: SamplingParams
     eos_token_id: Optional[int]
diff --git a/vllm/v1/engine/core.py b/vllm/v1/engine/core.py
index fdb241e6753fb..625e2d943666f 100644
--- a/vllm/v1/engine/core.py
+++ b/vllm/v1/engine/core.py
@@ -19,7 +19,7 @@
 from vllm.v1.engine import (EngineCoreOutput, EngineCoreOutputs,
                             EngineCoreProfile, EngineCoreRequest,
                             EngineCoreRequestType)
-from vllm.v1.engine.mm_input_mapper import MMInputMapper
+from vllm.v1.engine.mm_input_mapper import MMInputMapperServer
 from vllm.v1.executor.abstract import Executor
 from vllm.v1.request import Request, RequestStatus
 from vllm.v1.serial_utils import PickleEncoder
@@ -56,9 +56,6 @@ def __init__(
         vllm_config.cache_config.num_gpu_blocks = num_gpu_blocks
         vllm_config.cache_config.num_cpu_blocks = num_cpu_blocks
 
-        # Set up multimodal input mapper (e.g., convert PIL images to tensors).
-        self.mm_input_mapper = MMInputMapper(vllm_config.model_config)
-
         # Setup scheduler.
         self.scheduler = Scheduler(vllm_config.scheduler_config,
                                    vllm_config.cache_config,
@@ -66,6 +63,8 @@ def __init__(
 
         self._last_logging_time = time.time()
 
+        self.mm_input_mapper_server = MMInputMapperServer()
+
     def _initialize_kv_caches(self,
                               cache_config: CacheConfig) -> Tuple[int, int]:
         start = time.time()
@@ -89,7 +88,13 @@ def _initialize_kv_caches(self,
 
     def add_request(self, request: EngineCoreRequest):
         """Add request to the scheduler."""
+
+        if request.mm_hashes is not None:
+            request.mm_inputs = self.mm_input_mapper_server.process_inputs(
+                request.mm_inputs, request.mm_hashes)
+
         req = Request.from_engine_core_request(request)
+
         self.scheduler.add_request(req)
 
     def abort_requests(self, request_ids: List[str]):
diff --git a/vllm/v1/engine/mm_input_mapper.py b/vllm/v1/engine/mm_input_mapper.py
index 7ad6882b04520..68c476cbca07b 100644
--- a/vllm/v1/engine/mm_input_mapper.py
+++ b/vllm/v1/engine/mm_input_mapper.py
@@ -1,11 +1,35 @@
 from typing import Any, Dict, List, Optional
 
+import PIL
+from blake3 import blake3
+
 from vllm.config import ModelConfig
+from vllm.inputs import PromptType
+from vllm.logger import init_logger
 from vllm.multimodal import (MULTIMODAL_REGISTRY, MultiModalDataDict,
                              MultiModalKwargs, MultiModalRegistry)
+from vllm.v1.utils import LRUDictCache
+
+logger = init_logger(__name__)
+
+# The idea of MM preprocessor caching is based on having a client and a server,
+# where the client executes in the frontend process (=P0) and the server in the
+# core process (=P1).
+#
+# -- Client: Executes the MM mapper and performs caching of the results.
+# -- Server: Performs caching of the results
+#
+# The caching for both client and server is mirrored/similar, and this allows us
+# to avoid the serialization of "mm_inputs" (like pixel values) between
+# client (=P0) and server (=P1) processes.
 
+# Both Client and Server must use the same cache size
+# (to perform mirrored caching)
+# TODO: Tune the MM cache size
+MM_CACHE_SIZE = 256
 
-class MMInputMapper:
+
+class MMInputMapperClient:
 
     def __init__(
         self,
@@ -18,23 +42,133 @@ def __init__(
             model_config)
         self.mm_registry.init_mm_limits_per_prompt(model_config)
 
+        self.mm_cache = LRUDictCache(MM_CACHE_SIZE)
+
+        # Set to None to disable (TODO: Disable!)
+        self.mm_debug_cache_hit_ratio_steps = 32
+        self.mm_cache_hits = 0
+        self.mm_cache_misses = 0
+
+    def cache_hit_ratio(self, steps) -> float:
+        total_steps = self.mm_cache_hits + self.mm_cache_misses
+
+        if total_steps > 0 and total_steps % steps == 0:
+            logger.debug("MMInputMapper: cache_hit_ratio = %.2f",
+                         self.mm_cache_hits / total_steps)
+
     def process_inputs(
         self,
         mm_data: MultiModalDataDict,
+        mm_hashes: Optional[List[str]],
         mm_processor_kwargs: Optional[Dict[str, Any]],
+        precomputed_mm_inputs: Optional[List[MultiModalKwargs]],
     ) -> List[MultiModalKwargs]:
+        if precomputed_mm_inputs is None:
+            image_inputs = mm_data["image"]
+            if not isinstance(image_inputs, list):
+                image_inputs = [image_inputs]
+            num_inputs = len(image_inputs)
+        else:
+            num_inputs = len(precomputed_mm_inputs)
+
+        # Check if hash is enabled
+        use_hash = mm_hashes is not None
+        if use_hash:
+            assert num_inputs == len(
+                mm_hashes), "num_inputs = {} len(mm_hashes) = {}".format(
+                    num_inputs, len(mm_hashes))
+
+        # Process each image input separately, so that later we can schedule
+        # them in a fine-grained manner.
+        # Apply caching (if enabled) and reuse precomputed inputs (if provided)
+        ret_hashes = [] if use_hash else None
+        ret_inputs: List[MultiModalKwargs] = []
+        for input_id in range(num_inputs):
+            if self.mm_debug_cache_hit_ratio_steps is not None:
+                self.cache_hit_ratio(self.mm_debug_cache_hit_ratio_steps)
+
+            mm_hash = None
+            mm_input = None
+            if use_hash:
+                mm_hash = mm_hashes[input_id]
+                mm_input = self.mm_cache.get(mm_hash)
+
+            if mm_input is None:
+                self.mm_cache_misses += 1
+                if precomputed_mm_inputs is not None:
+                    # Reuse precomputed input (for merged preprocessor)
+                    mm_input = precomputed_mm_inputs[input_id]
+                else:
+                    # Apply MM mapper
+                    mm_input = self.multi_modal_input_mapper(
+                        {"image": [image_inputs[input_id]]},
+                        mm_processor_kwargs=mm_processor_kwargs,
+                    )
+
+                if use_hash:
+                    # Add to cache
+                    self.mm_cache.put(mm_hash, mm_input)
+            else:
+                self.mm_cache_hits += 1
+                mm_input = None  # Avoids sending mm_input to Server
+
+            if use_hash:
+                ret_hashes.append(mm_hash)
+            ret_inputs.append(mm_input)
+
+        return ret_inputs, ret_hashes
+
+
+class MMInputMapperServer:
+
+    def __init__(self, ):
+        self.mm_cache = LRUDictCache(MM_CACHE_SIZE)
+
+    def process_inputs(
+        self,
+        mm_inputs: List[Optional[MultiModalKwargs]],
+        mm_hashes: List[Optional[str]],
+    ) -> List[MultiModalKwargs]:
+        assert len(mm_inputs) == len(mm_hashes)
+
+        full_mm_inputs = []
+        for mm_input, mm_hash in zip(mm_inputs, mm_hashes):
+            if mm_input is None:
+                mm_input = self.mm_cache.get(mm_hash)
+                assert mm_input is not None
+            else:
+                self.mm_cache.put(mm_hash, mm_input)
+
+            full_mm_inputs.append(mm_input)
+
+        return full_mm_inputs
+
+
+class MMHasher:
+
+    def __init__(self):
+        pass
+
+    def hash(self, prompt: PromptType) -> Optional[List[str]]:
+        if "multi_modal_data" not in prompt:
+            return None
+
+        mm_data = prompt["multi_modal_data"]
         image_inputs = mm_data["image"]
         if not isinstance(image_inputs, list):
             image_inputs = [image_inputs]
+        assert len(image_inputs) > 0
 
-        # Process each image input separately so that later we can schedule
-        # them in a fine-grained manner.
-        mm_inputs: List[MultiModalKwargs] = []
-        num_images = len(image_inputs)
-        for i in range(num_images):
-            mm_input = self.multi_modal_input_mapper(
-                {"image": image_inputs[i]},
-                mm_processor_kwargs=mm_processor_kwargs,
-            )
-            mm_inputs.append(mm_input)
-        return mm_inputs
+        ret = []
+        for image in image_inputs:
+            assert isinstance(image, PIL.Image.Image)
+
+            # Convert image to bytes
+            bytes = image.tobytes()
+
+            # Hash image bytes
+            hasher = blake3()
+            hasher.update(bytes)
+            ret.append(hasher.hexdigest())
+
+        return ret
diff --git a/vllm/v1/engine/processor.py b/vllm/v1/engine/processor.py
index 120fc64969552..99ebb33303012 100644
--- a/vllm/v1/engine/processor.py
+++ b/vllm/v1/engine/processor.py
@@ -15,7 +15,7 @@
 from vllm.transformers_utils.config import try_get_generation_config
 from vllm.transformers_utils.tokenizer_group import BaseTokenizerGroup
 from vllm.v1.engine import DetokenizerRequest, EngineCoreRequest
-from vllm.v1.engine.mm_input_mapper import MMInputMapper
+from vllm.v1.engine.mm_input_mapper import MMHasher, MMInputMapperClient
 
 
 class Processor:
@@ -42,7 +42,11 @@ def __init__(
             model_config)
 
         # Multi-modal (huggingface) input mapper
-        self.mm_input_mapper = MMInputMapper(model_config)
+        self.mm_input_mapper_client = MMInputMapperClient(model_config)
+
+        # Multi-modal hasher (for images)
+        self.mm_hasher = MMHasher(
+        ) if model_config.mm_cache_preprocessor else None
 
     # TODO: run in an ThreadpoolExecutor or BackgroundProcess.
     # This ideally should releases the GIL, so we should not block the
@@ -71,6 +75,11 @@ def process_inputs(
         assert priority == 0, "vLLM V1 does not support priority at the moment."
         assert trace_headers is None, "vLLM V1 does not support tracing yet."
 
+        # Compute MM hashes (if enabled)
+        mm_hashes = None
+        if self.mm_hasher is not None:
+            mm_hashes = self.mm_hasher.hash(prompt)
+
         # Process inputs.
         preprocessed_inputs = self.input_preprocessor.preprocess(
             prompt,
@@ -101,16 +110,17 @@ def process_inputs(
         sampling_params.update_from_generation_config(
             self.generation_config_fields, eos_token_id)
 
-        # Preprocess multi-modal data
-        if len(decoder_inputs.multi_modal_data) == 0:
-            mm_inputs = None
-        elif isinstance(decoder_inputs.multi_modal_data, MultiModalKwargs):
-            mm_inputs = [decoder_inputs.multi_modal_data]
-        else:
-            mm_inputs = self.mm_input_mapper.process_inputs(
-                decoder_inputs.multi_modal_data,
-                decoder_inputs.mm_processor_kwargs,
-            )
+        # For merged preprocessor, mm_data is already mm_inputs
+        precomputed_mm_inputs = None
+        if isinstance(decoder_inputs.multi_modal_data, MultiModalKwargs):
+            precomputed_mm_inputs = [decoder_inputs.multi_modal_data]
+
+        # Apply MM mapper
+        mm_inputs = None
+        if len(decoder_inputs.multi_modal_data) > 0:
+            mm_inputs, mm_hashes = self.mm_input_mapper_client.process_inputs(
+                decoder_inputs.multi_modal_data, mm_hashes,
+                decoder_inputs.mm_processor_kwargs, precomputed_mm_inputs)
 
         # Make Request for Detokenizer.
         detokenizer_request = DetokenizerRequest(
@@ -130,6 +140,7 @@ def process_inputs(
             decoder_inputs.prompt,
             decoder_inputs.prompt_token_ids,
             mm_inputs,
+            mm_hashes,
             decoder_inputs.multi_modal_placeholders,
             sampling_params,
             eos_token_id,
diff --git a/vllm/v1/utils.py b/vllm/v1/utils.py
index 6e7a7d4fe12cd..d7911afee3ddb 100644
--- a/vllm/v1/utils.py
+++ b/vllm/v1/utils.py
@@ -1,5 +1,6 @@
 from contextlib import contextmanager
 from typing import Any, Generic, Iterator, List, TypeVar, overload
+from collections import OrderedDict
 
 import zmq
 
@@ -93,3 +94,23 @@ def make_zmq_socket(path: str, type: Any) -> Iterator[zmq.Socket]:
 
     finally:
         ctx.destroy(linger=0)
+
+
+class LRUDictCache:
+
+    def __init__(self, size: int):
+        self.cache = OrderedDict()
+        self.size = size
+
+    def get(self, key, default=None):
+        if key not in self.cache:
+            return default
+
+        self.cache.move_to_end(key)
+        return self.cache[key]
+
+    def put(self, key, value):
+        self.cache[key] = value
+        self.cache.move_to_end(key)
+        if len(self.cache) > self.size:
+            self.cache.popitem(last=False)