Sync from GitHub via hub-sync

.gitignore

@@ -1,6 +0,0 @@
-.DS_Store
-__pycache__/
-*.pyc
-.ruff_cache/
-.venv/
-*.log

CLAUDE.md

@@ -14,7 +14,11 @@ if not torch.cuda.is_available():
 ```
 
 ### 2. vLLM Docker Image
-Always use `vllm/vllm-openai:latest` for HF Jobs - it has all dependencies pre-installed.
+Run on the pinned **`vllm/vllm-openai:0.22.1`** image for HF Jobs (ships vLLM + the CUDA toolkit, so FlashInfer works) with `--python /usr/bin/python3 -e PYTHONPATH=/usr/local/lib/python3.12/dist-packages`. **Pin the tag (test-then-pin), not `:latest`.** Tested 2026-06-05 on `l4x1` (vLLM 0.22.1).
+
+The custom nightly / FlashInfer `[[tool.uv.index]]` blocks (§3) are **no longer needed** — the image provides vLLM; keep a `vllm>=0.11.0` floor in the PEP 723 header for the local path.
+
+**APIs (verified):** structured outputs use `StructuredOutputsParams` (`structured_outputs=`, ≥0.11.0 — `GuidedDecodingParams`/`guided_decoding=` removed in 0.12.0); classification auto-detects the pooling runner from the model architecture (the old `LLM(..., task="classify")` arg was removed → use `LLM(model)`).
 
 ### 3. Dependencies
 Include custom PyPI indexes for vLLM and FlashInfer:

classify-dataset.py

@@ -2,21 +2,14 @@
 # requires-python = ">=3.10"
 # dependencies = [
 #     "datasets",
-#     "flashinfer-python",
 #     "httpx",
-#     "huggingface-hub[hf_transfer]",
+#     "huggingface-hub",
 #     "setuptools",
 #     "toolz",
 #     "torch",
 #     "transformers",
-#     "vllm",
+#     "vllm>=0.11.0",
 # ]
-#
-# [[tool.uv.index]]
-# url = "https://flashinfer.ai/whl/cu126/torch2.6"
-#
-# [[tool.uv.index]]
-# url = "https://wheels.vllm.ai/nightly"
 # ///
 """
 Batch text classification using vLLM for efficient GPU inference.
@@ -155,9 +148,10 @@ def main(
         )
         sys.exit(1)
 
-    # Initialize vLLM with classification task
+    # vLLM auto-detects the sequence-classification runner from the model
+    # architecture (*ForSequenceClassification); the task= arg was removed in 0.12.0.
     logger.info(f"Loading model: {hub_model_id}")
-    llm = LLM(model=hub_model_id, task="classify")
+    llm = LLM(model=hub_model_id)
 
     # Get label mapping if available
     id2label = get_model_id2label(hub_model_id)
@@ -277,7 +271,7 @@ This script requires arguments. For usage information:
 Example HF Jobs command:
     hfjobs run \\
         --flavor l4x1 \\
-        --secret HF_TOKEN=\$(python -c "from huggingface_hub import HfFolder; print(HfFolder.get_token())") \\
+        --secret HF_TOKEN=$(python -c "from huggingface_hub import HfFolder; print(HfFolder.get_token())") \\
         vllm/vllm-openai:latest \\
         /bin/bash -c '
             uv run https://huggingface.co/datasets/uv-scripts/vllm/resolve/main/classify-dataset.py \\

generate-responses.py

@@ -3,11 +3,11 @@
 # dependencies = [
 #     "datasets",
 #     "flashinfer-python",
-#     "huggingface-hub[hf_transfer]",
+#     "huggingface-hub",
 #     "hf-xet>= 1.1.7",
 #     "torch",
 #     "transformers",
-#     "vllm>=0.8.5",
+#     "vllm>=0.11.0",
 # ]
 #
 # ///
@@ -52,9 +52,6 @@ from tqdm.auto import tqdm
 from transformers import AutoTokenizer
 from vllm import LLM, SamplingParams
 
-# Enable HF Transfer for faster downloads
-os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
-
 logging.basicConfig(
     level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
 )

vlm-classify.py

@@ -2,13 +2,13 @@
 # requires-python = ">=3.11"
 # dependencies = [
 #     "datasets",
-#     "huggingface-hub[hf_transfer]",
+#     "huggingface-hub",
 #     "pillow",
 #     "toolz",
 #     "torch",
 #     "tqdm",
 #     "transformers",
-#     "vllm>=0.6.5",
+#     "vllm>=0.11.0",  # StructuredOutputsParams API (guided_decoding removed in 0.12.0)
 # ]
 # ///
 
@@ -52,12 +52,12 @@ from typing import List, Optional, Union, Dict, Any
 
 import torch
 from PIL import Image
-from datasets import load_dataset, Dataset
+from datasets import load_dataset
 from huggingface_hub import login
 from toolz import partition_all
 from tqdm.auto import tqdm
 from vllm import LLM, SamplingParams
-from vllm.sampling_params import GuidedDecodingParams
+from vllm.sampling_params import StructuredOutputsParams
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
@@ -194,12 +194,14 @@ def main(
 
     llm = LLM(**llm_kwargs)
 
-    # Create guided decoding params for classification
-    guided_decoding_params = GuidedDecodingParams(choice=class_list)
+    # Constrain output to one of the class labels (vLLM structured outputs).
+    # StructuredOutputsParams + structured_outputs= is the API from vLLM 0.11.0+;
+    # the old GuidedDecodingParams/guided_decoding= was removed in 0.12.0.
+    structured = StructuredOutputsParams(choice=class_list)
     sampling_params = SamplingParams(
         temperature=0.1,  # Low temperature for consistent classification
         max_tokens=50,  # Classifications are short
-        guided_decoding=guided_decoding_params,
+        structured_outputs=structured,
     )
 
     # Process images in batches to avoid memory issues