Sync from GitHub via hub-sync

README.md

@@ -34,7 +34,7 @@ The script name says `qwen3vl` because the bbox-output convention (`bbox_2d` JSO
 
 ```bash
 hf jobs uv run \
-  --image vllm/vllm-openai:latest \
+  --image vllm/vllm-openai:v0.32.0 \
   --flavor a100-large \
   --python /usr/bin/python3 \
   -e PYTHONPATH=/usr/local/lib/python3.12/dist-packages \
@@ -121,13 +121,14 @@ not return overlapping or duplicate boxes.
 
 ## Hardware notes
 
-- **A100-large + `vllm/vllm-openai:latest` image** is the working combo. The default uv-script image installs PyPI's latest PyTorch, which has a CUDA driver mismatch on a100 hosts (driver 12.9, PyTorch wants newer). Workaround: `--image vllm/vllm-openai:latest --python /usr/bin/python3 -e PYTHONPATH=/usr/local/lib/python3.12/dist-packages` (uses the image's pre-installed driver-compatible torch + vllm).
+- **A100-large + `vllm/vllm-openai:v0.32.0` image** is the working combo. The default uv-script image installs PyPI's latest PyTorch, which has a CUDA driver mismatch on a100 hosts (driver 12.9, PyTorch wants newer). Workaround: `--image vllm/vllm-openai:v0.32.0 --python /usr/bin/python3 -e PYTHONPATH=/usr/local/lib/python3.12/dist-packages` (uses the image's pre-installed driver-compatible torch + vllm).
+- **Local GPU runs**: the script headers leave `vllm`, `torch` and `transformers` to the Jobs image, so on your own Linux + CUDA machine add vLLM explicitly: `uv run --with vllm qwen3vl-detect.py …`.
 - **H200 was unreliable** in our testing — flashinfer's runtime JIT compile of cutlass kernels for `compute_90a` failed in the same image. Stick with A100 for now.
 - **`max_num_seqs` is auto-capped to `batch_size`**. Qwen3.6's hybrid Gated-DeltaNet architecture allocates per-sequence SSM cache blocks; vLLM's default 256 can exceed available blocks and crash CUDA graph capture. We never decode > batch_size concurrently anyway.
 
 ## CLI reference
 
-Run `uv run qwen3vl-detect.py --help` for the full flag set. Most-tuned flags:
+Run `uv run qwen3vl-detect.py --help` for the full flag set (no GPU or vLLM needed). Most-tuned flags:
 
 - `--prompt` / `--prompt-file` — the detection instruction
 - `--model` — any vLLM-supported VLM. Default: `Qwen/Qwen3.6-35B-A3B`. For smaller GPUs use `Qwen/Qwen3.5-9B` and lower `--max-model-len` / `--max-image-tokens`.

qwen3vl-detect-fewshot.py

@@ -5,13 +5,14 @@
 #     "huggingface-hub",
 #     "pillow",
 #     "toolz",
-#     "torch",
 #     "tqdm",
-#     "transformers",
-#     "vllm>=0.15.1",
 # ]
 #
 # [tool.hf-jobs]
+# # vllm, torch and transformers come from the image (PYTHONPATH). Listing them above makes uv
+# # install a newer PyPI vllm and its deps; PYTHONPATH shadows that vllm but not new deps such
+# # as b12x, which the image's vllm then imports and crashes on (missing flashinfer module).
+# # Seen on v0.32.0 (2026-11-05); re-test before moving the tag. Local GPU: uv run --with vllm.
 # image = "vllm/vllm-openai:v0.32.0"
 # python = "/usr/bin/python3"
 # env = { PYTHONPATH = "/usr/local/lib/python3.12/dist-packages" }
@@ -59,14 +60,11 @@ import sys
 from datetime import datetime
 from typing import Any, Optional, Union
 
-import torch
 from datasets import load_dataset
 from huggingface_hub import login
 from PIL import Image
 from toolz import partition_all
 from tqdm.auto import tqdm
-from transformers import AutoProcessor
-from vllm import LLM, SamplingParams
 
 logging.basicConfig(
     level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s"
@@ -221,6 +219,12 @@ def main(
     seed: int = 42,
     create_pr: bool = False,
 ) -> None:
+    # Imported here rather than at module top so `uv run <script> --help` works without
+    # them; on Jobs they come from the vLLM image (see the header).
+    import torch
+    from transformers import AutoProcessor
+    from vllm import LLM, SamplingParams
+
     if not torch.cuda.is_available():
         logger.error("CUDA is not available. This script requires a GPU.")
         logger.error("For cloud execution: hf jobs uv run --flavor a100-large ...")
@@ -456,13 +460,15 @@ def build_parser() -> argparse.ArgumentParser:
         formatter_class=argparse.RawDescriptionHelpFormatter,
         epilog="""
 Examples:
+  # Local GPU (Linux + CUDA): the header leaves vllm/torch/transformers to
+  # the Jobs image, so add vllm with --with.
   # Generic detection
-  uv run qwen3vl-detect.py \\
+  uv run --with vllm qwen3vl-detect-fewshot.py \\
       input-dataset username/output-dataset \\
       --prompt "Detect all visible objects, return JSON with bbox_2d, label, sub_label"
 
   # Component detection (matches the demo's example 1)
-  uv run qwen3vl-detect.py \\
+  uv run --with vllm qwen3vl-detect-fewshot.py \\
       pcb-images username/labelled-pcbs \\
       --prompt-file prompts/components.txt \\
       --max-samples 20

qwen3vl-detect.py

@@ -5,13 +5,14 @@
 #     "huggingface-hub",
 #     "pillow",
 #     "toolz",
-#     "torch",
 #     "tqdm",
-#     "transformers",
-#     "vllm>=0.15.1",
 # ]
 #
 # [tool.hf-jobs]
+# # vllm, torch and transformers come from the image (PYTHONPATH). Listing them above makes uv
+# # install a newer PyPI vllm and its deps; PYTHONPATH shadows that vllm but not new deps such
+# # as b12x, which the image's vllm then imports and crashes on (missing flashinfer module).
+# # Seen on v0.32.0 (2026-11-05); re-test before moving the tag. Local GPU: uv run --with vllm.
 # image = "vllm/vllm-openai:v0.32.0"
 # python = "/usr/bin/python3"
 # env = { PYTHONPATH = "/usr/local/lib/python3.12/dist-packages" }
@@ -53,14 +54,11 @@ import sys
 from datetime import datetime
 from typing import Any, Optional, Union
 
-import torch
 from datasets import load_dataset
 from huggingface_hub import login
 from PIL import Image
 from toolz import partition_all
 from tqdm.auto import tqdm
-from transformers import AutoProcessor
-from vllm import LLM, SamplingParams
 
 logging.basicConfig(
     level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s"
@@ -213,6 +211,12 @@ def main(
     seed: int = 42,
     create_pr: bool = False,
 ) -> None:
+    # Imported here rather than at module top so `uv run <script> --help` works without
+    # them; on Jobs they come from the vLLM image (see the header).
+    import torch
+    from transformers import AutoProcessor
+    from vllm import LLM, SamplingParams
+
     if not torch.cuda.is_available():
         logger.error("CUDA is not available. This script requires a GPU.")
         logger.error("For cloud execution: hf jobs uv run --flavor a100-large ...")
@@ -407,13 +411,15 @@ def build_parser() -> argparse.ArgumentParser:
         formatter_class=argparse.RawDescriptionHelpFormatter,
         epilog="""
 Examples:
+  # Local GPU (Linux + CUDA): the header leaves vllm/torch/transformers to
+  # the Jobs image, so add vllm with --with.
   # Generic detection
-  uv run qwen3vl-detect.py \\
+  uv run --with vllm qwen3vl-detect.py \\
       input-dataset username/output-dataset \\
       --prompt "Detect all visible objects, return JSON with bbox_2d, label, sub_label"
 
   # Component detection (matches the demo's example 1)
-  uv run qwen3vl-detect.py \\
+  uv run --with vllm qwen3vl-detect.py \\
       pcb-images username/labelled-pcbs \\
       --prompt-file prompts/components.txt \\
       --max-samples 20