[ray-update-2.56.0] Update llm_batch_inference_vision to Ray 2.56.0

BUILD.yaml

@@ -589,7 +589,7 @@
   owner_team: llm
   dir: templates/llm_batch_inference_vision
   cluster_env:
-    image_uri: anyscale/ray-llm:2.55.1-py311-cu128
+    image_uri: anyscale/ray-llm:2.56.0-py312-cu130
   compute_config:
     AWS: configs/llm_batch_inference_vision/aws.yaml
     GCP: configs/llm_batch_inference_vision/gce.yaml

dependencies/template.depsets.yaml

@@ -177,12 +177,12 @@ depsets:
       - templates/llm_batch_inference_vision/requirements.txt
     output: templates/llm_batch_inference_vision/python_depset.lock
     append_flags:
-      - --index https://download.pytorch.org/whl/cu128
+      - --index https://download.pytorch.org/whl/cu130
       - --python-version=${PYTHON_VERSION}
       - --python-platform=x86_64-manylinux_2_31
       - --unsafe-package ray
     build_arg_sets:
-      - ray2551_py311_cu128
+      - ray2560_py312_cu130
 #
 #  - name: llm_finetuning_depset_${RAY_VERSION}_${PYTHON_VERSION}_${CUDA_VARIANT}
 #    operation: expand

templates/llm_batch_inference_vision/README.ipynb

@@ -274,10 +274,10 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from ray.data.llm import build_llm_processor\n",
+    "from ray.data.llm import build_processor\n",
     "\n",
     "# Build the LLM processor with the configuration and functions.\n",
-    "processor = build_llm_processor(\n",
+    "processor = build_processor(\n",
     "    processor_config,\n",
     "    preprocess=preprocess,\n",
     "    postprocess=postprocess,\n",
@@ -339,7 +339,7 @@
     "# job.yaml\n",
     "name: my-llm-batch-inference-vision\n",
     "entrypoint: python batch_inference_vision.py\n",
-    "image_uri: anyscale/ray-llm:2.55.1-py311-cu128\n",
+    "image_uri: anyscale/ray-llm:2.56.0-py312-cu130\n",
     "compute_config:\n",
     "  head_node:\n",
     "    instance_type: m5.2xlarge\n",
@@ -455,7 +455,7 @@
     ")\n",
     "\n",
     "# Build the LLM processor with the configuration and functions.\n",
-    "processor_large = build_llm_processor(\n",
+    "processor_large = build_processor(\n",
     "    processor_config_large,\n",
     "    preprocess=preprocess,\n",
     "    postprocess=postprocess,\n",

templates/llm_batch_inference_vision/README.md

@@ -209,10 +209,10 @@ With the configuration and functions defined, build the processor.
 
 
 ```python
-from ray.data.llm import build_llm_processor
+from ray.data.llm import build_processor
 
 # Build the LLM processor with the configuration and functions.
-processor = build_llm_processor(
+processor = build_processor(
     processor_config,
     preprocess=preprocess,
     postprocess=postprocess,
@@ -262,7 +262,7 @@ Save your batch inference code as `batch_inference_vision.py`, then create a job
 # job.yaml
 name: my-llm-batch-inference-vision
 entrypoint: python batch_inference_vision.py
-image_uri: anyscale/ray-llm:2.55.1-py311-cu128
+image_uri: anyscale/ray-llm:2.56.0-py312-cu130
 compute_config:
   head_node:
     instance_type: m5.2xlarge
@@ -347,7 +347,7 @@ processor_config_large = vLLMEngineProcessorConfig(
 )
 
 # Build the LLM processor with the configuration and functions.
-processor_large = build_llm_processor(
+processor_large = build_processor(
     processor_config_large,
     preprocess=preprocess,
     postprocess=postprocess,

templates/llm_batch_inference_vision/batch_inference_vision.py

@@ -5,7 +5,7 @@
 from PIL import Image
 from pprint import pprint
 import ray
-from ray.data.llm import build_llm_processor, vLLMEngineProcessorConfig
+from ray.data.llm import build_processor, vLLMEngineProcessorConfig
 
 DATASET_LIMIT = 10_000
 
@@ -99,7 +99,7 @@ def postprocess(row: dict[str, Any]) -> dict[str, Any]:
     }
 
 # Build the LLM processor with the configuration and functions.
-processor = build_llm_processor(
+processor = build_processor(
     processor_config,
     preprocess=preprocess,
     postprocess=postprocess,

templates/llm_batch_inference_vision/batch_inference_vision_scaled.py

@@ -5,7 +5,7 @@
 from PIL import Image
 from pprint import pprint
 import ray
-from ray.data.llm import build_llm_processor, vLLMEngineProcessorConfig
+from ray.data.llm import build_processor, vLLMEngineProcessorConfig
 
 # Dataset limit for this example.
 DATASET_LIMIT = 1_000_000
@@ -94,7 +94,7 @@ def postprocess(row: dict[str, Any]) -> dict[str, Any]:
     }
 
 # Build the LLM processor with the configuration and functions.
-processor_large = build_llm_processor(
+processor_large = build_processor(
     processor_config_large,
     preprocess=preprocess,
     postprocess=postprocess,

templates/llm_batch_inference_vision/job.yaml

@@ -1,6 +1,6 @@
 name: llm-batch-inference-vision
 entrypoint: python batch_inference_vision.py
-image_uri: anyscale/ray-llm:2.55.1-py311-cu128
+image_uri: anyscale/ray-llm:2.56.0-py312-cu130
 compute_config:
   head_node:
     instance_type: m5.2xlarge

templates/llm_batch_inference_vision/python_depset.lock

@@ -1,5 +1,5 @@
 --index-url https://pypi.org/simple
---extra-index-url https://download.pytorch.org/whl/cu128
+--extra-index-url https://download.pytorch.org/whl/cu130
 
 aiohappyeyeballs==2.6.2 \
     --hash=sha256:4708045e2d7a6c6bdf8aafa8ed39649eaf926a4543b54560659129e3365953c4 \
@@ -131,14 +131,25 @@ aiosignal==1.4.0 \
     --hash=sha256:053243f8b92b990551949e63930a839ff0cf0b0ebbe0597b0f3fb19e1a0fe82e \
     --hash=sha256:f47eecd9468083c2029cc99945502cb7708b082c232f9aca65da147157b251c7
     # via aiohttp
+annotated-doc==0.0.4 \
+    --hash=sha256:571ac1dc6991c450b25a9c2d84a3705e2ae7a53467b5d111c24fa8baabbed320 \
+    --hash=sha256:fbcda96e87e9c92ad167c2e53839e57503ecfda18804ea28102353485033faa4
+    # via typer
+anyio==4.14.1 \
+    --hash=sha256:4e5533c5b8ff0a24f5d7a176cbe6877129cd183893f66b537f8f227d10527d72 \
+    --hash=sha256:8d648a3544c1a700e3ff78615cd679e4c5c3f149904287e73687b2596963629e
+    # via httpx
 attrs==26.1.0 \
     --hash=sha256:c647aa4a12dfbad9333ca4e71fe62ddc36f4e63b2d260a37a8b83d2f043ac309 \
     --hash=sha256:d03ceb89cb322a8fd706d4fb91940737b6642aa36998fe130a9bc96c985eff32
     # via aiohttp
 certifi==2026.5.20 \
     --hash=sha256:3c52e209ba0a4ad7aebe60436a4ab349c39e1e602e8c134221e546902ad25897 \
     --hash=sha256:69dea482ab64caa7b9f6aba1c6bf48bb6a5448d1c0f1b17ab42ad8c763a5344d
-    # via requests
+    # via
+    #   httpcore
+    #   httpx
+    #   requests
 charset-normalizer==3.4.7 \
     --hash=sha256:007d05ec7321d12a40227aae9e2bc6dca73f3cb21058999a1df9e193555a9dcc \
     --hash=sha256:03853ed82eeebbce3c2abfdbc98c96dc205f32a79627688ac9a27370ea61a49c \
@@ -426,7 +437,11 @@ fsspec==2025.3.0 \
     # via
     #   datasets
     #   huggingface-hub
-hf-xet==1.5.0 ; platform_machine == 'aarch64' or platform_machine == 'amd64' or platform_machine == 'arm64' or platform_machine == 'x86_64' \
+h11==0.16.0 \
+    --hash=sha256:4e35b956cf45792e4caa5885e69fba00bdbc6ffafbfa020300e549b208ee5ff1 \
+    --hash=sha256:63cf8bbe7522de3bf65932fda1d9c2772064ffb3dae62d55932da54b31cb6c86
+    # via httpcore
+hf-xet==1.5.0 ; platform_machine == 'AMD64' or platform_machine == 'aarch64' or platform_machine == 'amd64' or platform_machine == 'arm64' or platform_machine == 'x86_64' \
     --hash=sha256:1e60df5a42e9bed8628b6416af2cba4cba57ae9f02de226a06b020d98e1aab18 \
     --hash=sha256:2806c7c17b4d23f8d88f7c4814f838c3b6150773fe339c20af23e1cfaf2797e4 \
     --hash=sha256:2baea1b0b989e5c152fe81425f7745ddc8901280ba3d97c98d8cdece7b706c60 \
@@ -453,18 +468,36 @@ hf-xet==1.5.0 ; platform_machine == 'aarch64' or platform_machine == 'amd64' or
     --hash=sha256:f7b7bbae318e583a86fb21e5a4a175d6721d628a2874f4bd022d0e660c32a682 \
     --hash=sha256:fd6e5a9b0fdac4ed03ed45ef79254a655b1aaab514a02202617fbf643f5fdf7a
     # via huggingface-hub
-huggingface-hub==0.36.2 \
-    --hash=sha256:1934304d2fb224f8afa3b87007d58501acfda9215b334eed53072dd5e815ff7a \
-    --hash=sha256:48f0c8eac16145dfce371e9d2d7772854a4f591bcb56c9cf548accf531d54270
+httpcore==1.0.9 \
+    --hash=sha256:2d400746a40668fc9dec9810239072b40b4484b640a8c38fd654a024c7a1bf55 \
+    --hash=sha256:6e34463af53fd2ab5d807f399a9b45ea31c3dfa2276f15a2c3f00afff6e176e8
+    # via httpx
+httpx==0.28.1 \
+    --hash=sha256:75e98c5f16b0f35b567856f597f06ff2270a374470a5c2392242528e3e3e42fc \
+    --hash=sha256:d909fcccc110f8c7faf814ca82a9a4d816bc5a6dbfea25d6591d6985b8ba59ad
+    # via huggingface-hub
+huggingface-hub==1.13.0 \
+    --hash=sha256:e942cb50d6a08dd5306688b1ac05bda157fd2fcc88b63dae405f7bd0d3234005 \
+    --hash=sha256:f6df2dac5abe82ce2fe05873d10d5ff47bc677d616a2f521f4ee26db9415d9d0
     # via
     #   -r templates/llm_batch_inference_vision/requirements.txt
     #   datasets
 idna==3.17 \
     --hash=sha256:466e48829084efe2548012b855df21540b96f2e20e51bd124c851536556a592c \
     --hash=sha256:5eb0cb53bc467c12eadcf6de83163ad8527cec9416f44b9b61b19caedad2b87f
     # via
+    #   anyio
+    #   httpx
     #   requests
     #   yarl
+markdown-it-py==4.2.0 \
+    --hash=sha256:04a21681d6fbb623de53f6f364d352309d4094dd4194040a10fd51833e418d49 \
+    --hash=sha256:9f7ebbcd14fe59494226453aed97c1070d83f8d24b6fc3a3bcf9a38092641c4a
+    # via rich
+mdurl==0.1.2 \
+    --hash=sha256:84008a41e51615a49fc9966191ff91509e3c40b939176e643fd50a5c2196b8f8 \
+    --hash=sha256:bb413d29f5eea38f31dd4754dd7377d4465116fb207585f97bf925588687c1ba
+    # via markdown-it-py
 multidict==6.7.1 \
     --hash=sha256:026d264228bcd637d4e060844e39cdc60f86c479e463d49075dedc21b18fbbe0 \
     --hash=sha256:03ede2a6ffbe8ef936b92cb4529f27f42be7f56afcdab5ab739cd5f27fb1cbf9 \
@@ -906,6 +939,10 @@ pyarrow==19.0.1 \
     # via
     #   -r templates/llm_batch_inference_vision/requirements.txt
     #   datasets
+pygments==2.20.0 \
+    --hash=sha256:6757cd03768053ff99f3039c1a36d6c0aa0b263438fcab17520b30a303a82b5f \
+    --hash=sha256:81a9e26dd42fd28a23a2d169d86d7ac03b46e2f8b59ed4698fb4785f946d0176
+    # via rich
 python-dateutil==2.9.0.post0 \
     --hash=sha256:37dd54208da7e1cd875388217d5e00ebd4179249f90fb72437e91a35459a0ad3 \
     --hash=sha256:a8b2bc7bffae282281c8140a97d3aa9c14da0b136dfe83f850eea9a5f7470427
@@ -994,9 +1031,15 @@ pyyaml==6.0.3 \
 requests==2.34.2 \
     --hash=sha256:2a0d60c172f83ac6ab31e4554906c0f3b3588d37b5cb939b1c061f4907e278e0 \
     --hash=sha256:f288924cae4e29463698d6d60bc6a4da69c89185ad1e0bcc4104f584e960b9ed
-    # via
-    #   datasets
-    #   huggingface-hub
+    # via datasets
+rich==15.0.0 \
+    --hash=sha256:33bd4ef74232fb73fe9279a257718407f169c09b78a87ad3d296f548e27de0bb \
+    --hash=sha256:edd07a4824c6b40189fb7ac9bc4c52536e9780fbbfbddf6f1e2502c31b068c36
+    # via typer
+shellingham==1.5.4 \
+    --hash=sha256:7ecfff8f2fd72616f7481040475a65b2bf8af90a56c89140852d1120324e8686 \
+    --hash=sha256:8dbca0739d487e5bd35ab3ca4b36e11c4078f3a234bfce294b0a0291363404de
+    # via typer
 six==1.17.0 \
     --hash=sha256:4721f391ed90541fddacab5acf947aa0d3dc7d27b2e1e8eda2be8970586c3274 \
     --hash=sha256:ff70335d468e7eb6ec65b95b99d3a2836546063f63acc5171de367e834932a81
@@ -1007,10 +1050,15 @@ tqdm==4.67.3 \
     # via
     #   datasets
     #   huggingface-hub
+typer==0.26.8 \
+    --hash=sha256:3512ca79ac5c11113414b36e80281b872884477722440691c89d1112e321a49c \
+    --hash=sha256:c244a6bd558886fe3f8780efb6bdd28bb9aff005a94eedebaa5cb32926fe2f7e
+    # via huggingface-hub
 typing-extensions==4.15.0 \
     --hash=sha256:f0fa19c6845758ab08074a0cfa8b7aecb71c999ca73d62883bc25cc018c4e548
     # via
     #   aiosignal
+    #   anyio
     #   huggingface-hub
 tzdata==2026.2 \
     --hash=sha256:9173fde7d80d9018e02a662e168e5a2d04f87c41ea174b139fbef642eda62d10 \

templates/llm_batch_inference_vision/requirements.txt

@@ -1,14 +1,15 @@
 # datasets 4.x requires the numpy-2 stack (numpy>=2, pyarrow>=21, pandas 3), which is
-# incompatible with the numpy-1.x base image (anyscale/ray-llm:2.55.1). The test layers
+# incompatible with the numpy-1.x base image (anyscale/ray-llm:2.56.0 still ships
+# numpy 1.26.4 + pyarrow 19.0.1 + scipy compiled against numpy 1.x). The test layers
 # this lock on the stock image with `uv pip install --no-deps`, so we keep the delta
 # purely additive: 3.6.0 is the newest datasets that resolves against the image's
 # numpy/pandas/pyarrow, which we pin so they reinstall as no-ops (no ABI break, Ray-safe).
 datasets==3.6.0
 numpy==1.26.4
 pandas==2.3.3
 pyarrow==19.0.1
-# datasets only requires huggingface-hub>=0.24.0, but unsafe-best-match resolves it to
-# 1.x; the base image's transformers==4.57.6 requires huggingface-hub<1.0, so a 1.x
-# upgrade breaks transformers import (the test layers this lock with --no-deps). Pin to
-# the base-image version so it reinstalls as a no-op.
-huggingface-hub==0.36.2
+# datasets only requires huggingface-hub>=0.24.0, but unsafe-best-match resolves to
+# a newer version than the image ships. Pin to the base-image version so it
+# reinstalls as a no-op (the 2.56.0 ray-llm image ships hf-hub 1.13.0 alongside
+# transformers 5.x).
+huggingface-hub==1.13.0