caption-flow 0.5.7__tar.gz → 0.6.1__tar.gz

{caption_flow-0.5.7 → caption_flow-0.6.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: caption-flow
-Version: 0.5.7
+Version: 0.6.1
 Summary: Self-contained distributed community captioning system
 Author: bghira
 Author-email: bghira <bghira@users.github.com>
@@ -16,82 +16,83 @@ Classifier: Programming Language :: Python :: 3.13
 Requires-Python: >=3.12,<3.14
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: websockets<17.0,>=16.0
-Requires-Dist: pyarrow<26.0.0,>=21.0.0
-Requires-Dist: click<9.0.0,>=8.2.0
-Requires-Dist: pydantic<3.0.0,>=2.12.0
-Requires-Dist: aiofiles<26.0.0,>=24.1.0
-Requires-Dist: rich<16.0.0,>=14.0.0
-Requires-Dist: cryptography<50.0.0,>=45.0.0
-Requires-Dist: PyYAML<7.0.0,>=6.0.2
-Requires-Dist: certbot<6.0.0,>=5.0.0
-Requires-Dist: numpy<3.0.0,>=2.2.0
-Requires-Dist: Pillow<13.0.0,>=11.3.0
-Requires-Dist: trainingsample<0.4.0,>=0.3.2
-Requires-Dist: pandas<4.0.0,>=3.0.0
-Requires-Dist: datasets<6.0.0,>=5.0.0
-Requires-Dist: boto3<2.0.0,>=1.43.0
-Requires-Dist: webshart<0.7.0,>=0.6.0
-Requires-Dist: pylance<9.0.0,>=8.0.0
-Requires-Dist: duckdb<2.0.0,>=1.5.0
-Requires-Dist: aiohttp<4.0.0,>=3.13.3
-Requires-Dist: fastapi<0.137.0,>=0.133.0
-Requires-Dist: uvicorn<1.0.0,>=0.35.0
-Requires-Dist: huggingface-hub<2.0.0,>=1.5.0
-Requires-Dist: psutil<8.0.0,>=7.0.0
-Requires-Dist: requests<3.0.0,>=2.32.0
-Requires-Dist: tqdm<5.0.0,>=4.67.0
-Requires-Dist: urwid<5.0.0,>=3.0.2
+Requires-Dist: websockets<18.0,>=17.4
+Requires-Dist: pyarrow<27.0.0,>=26.0.1
+Requires-Dist: click<9.0.0,>=8.5.1
+Requires-Dist: pydantic<3.0.0,>=2.14.3
+Requires-Dist: aiofiles<26.0.0,>=25.1.0
+Requires-Dist: rich<16.0.0,>=15.0.0
+Requires-Dist: cryptography<51.0.0,>=50.0.2
+Requires-Dist: PyYAML<7.0.0,>=6.0.3
+Requires-Dist: certbot<6.0.0,>=5.10.0
+Requires-Dist: numpy<2.4.0,>=2.3.5; python_version < "3.13"
+Requires-Dist: numpy<2.5.0,>=2.4.6; python_version >= "3.13"
+Requires-Dist: Pillow<14.0.0,>=13.0.0
+Requires-Dist: trainingsample==0.4.0
+Requires-Dist: pandas<4.0.0,>=3.1.1
+Requires-Dist: datasets<6.0.0,>=5.2.0
+Requires-Dist: boto3<2.0.0,>=1.44.3
+Requires-Dist: webshart<0.8.0,>=0.7.0
+Requires-Dist: pylance<9.0.0,>=8.0.1
+Requires-Dist: duckdb<2.0.0,>=1.5.6
+Requires-Dist: aiohttp<4.0.0,>=3.14.5
+Requires-Dist: fastapi<0.137.0,>=0.136.3
+Requires-Dist: uvicorn<1.0.0,>=0.54.0
+Requires-Dist: huggingface-hub<2.0.0,>=1.33.0
+Requires-Dist: psutil<8.0.0,>=7.2.2
+Requires-Dist: requests<3.0.0,>=2.34.2
+Requires-Dist: tqdm<5.0.0,>=4.70.1
+Requires-Dist: urwid<6.0.0,>=5.1.3
 Provides-Extra: vllm
-Requires-Dist: vllm<0.32.0,>=0.31.0; extra == "vllm"
-Requires-Dist: torch==2.13.0; extra == "vllm"
-Requires-Dist: torchvision==0.28.0; extra == "vllm"
+Requires-Dist: vllm<0.34.0,>=0.33.0; extra == "vllm"
+Requires-Dist: torch==2.14.0; extra == "vllm"
+Requires-Dist: torchvision==0.29.0; extra == "vllm"
 Requires-Dist: torchaudio==2.11.0; extra == "vllm"
-Requires-Dist: transformers<5.18.0,>=5.10.4; extra == "vllm"
+Requires-Dist: transformers<5.21.0,>=5.16.1; extra == "vllm"
 Requires-Dist: qwen-vl-utils<0.1.0,>=0.0.14; extra == "vllm"
 Provides-Extra: captioning
-Requires-Dist: vllm<0.32.0,>=0.31.0; extra == "captioning"
-Requires-Dist: torch==2.13.0; extra == "captioning"
-Requires-Dist: torchvision==0.28.0; extra == "captioning"
+Requires-Dist: vllm<0.34.0,>=0.33.0; extra == "captioning"
+Requires-Dist: torch==2.14.0; extra == "captioning"
+Requires-Dist: torchvision==0.29.0; extra == "captioning"
 Requires-Dist: torchaudio==2.11.0; extra == "captioning"
-Requires-Dist: transformers<5.18.0,>=5.10.4; extra == "captioning"
+Requires-Dist: transformers<5.21.0,>=5.16.1; extra == "captioning"
 Requires-Dist: qwen-vl-utils<0.1.0,>=0.0.14; extra == "captioning"
 Provides-Extra: openai
 Provides-Extra: cpu
-Requires-Dist: torch>=2.13.0; extra == "cpu"
-Requires-Dist: torchvision>=0.28.0; extra == "cpu"
+Requires-Dist: torch>=2.14.0; extra == "cpu"
+Requires-Dist: torchvision>=0.29.0; extra == "cpu"
 Requires-Dist: torchaudio>=2.11.0; extra == "cpu"
 Provides-Extra: cuda
-Requires-Dist: torch>=2.13.0; extra == "cuda"
-Requires-Dist: torchvision>=0.28.0; extra == "cuda"
+Requires-Dist: torch>=2.14.0; extra == "cuda"
+Requires-Dist: torchvision>=0.29.0; extra == "cuda"
 Requires-Dist: torchaudio>=2.11.0; extra == "cuda"
 Provides-Extra: cuda13
-Requires-Dist: torch>=2.13.0; extra == "cuda13"
-Requires-Dist: torchvision>=0.28.0; extra == "cuda13"
+Requires-Dist: torch>=2.14.0; extra == "cuda13"
+Requires-Dist: torchvision>=0.29.0; extra == "cuda13"
 Requires-Dist: torchaudio>=2.11.0; extra == "cuda13"
 Provides-Extra: rocm
-Requires-Dist: torch>=2.13.0; extra == "rocm"
-Requires-Dist: torchvision>=0.28.0; extra == "rocm"
+Requires-Dist: torch>=2.14.0; extra == "rocm"
+Requires-Dist: torchvision>=0.29.0; extra == "rocm"
 Requires-Dist: torchaudio>=2.11.0; extra == "rocm"
 Provides-Extra: apple
-Requires-Dist: torch>=2.13.0; extra == "apple"
-Requires-Dist: torchvision>=0.28.0; extra == "apple"
+Requires-Dist: torch>=2.14.0; extra == "apple"
+Requires-Dist: torchvision>=0.29.0; extra == "apple"
 Requires-Dist: torchaudio>=2.11.0; extra == "apple"
-Requires-Dist: vllm-metal==0.1.0; (platform_system == "Darwin" and platform_machine == "arm64" and python_version >= "3.12") and extra == "apple"
+Requires-Dist: qwen-vl-utils<0.1.0,>=0.0.14; extra == "apple"
 Provides-Extra: dev
-Requires-Dist: pytest<9.0.0,>=8.0.0; extra == "dev"
-Requires-Dist: pytest-asyncio<2.0.0,>=1.1.0; extra == "dev"
-Requires-Dist: pytest-cov<7.0.0,>=6.0.0; extra == "dev"
-Requires-Dist: transformers<5.18.0,>=5.10.4; extra == "dev"
-Requires-Dist: black<26.0.0,>=25.0.0; extra == "dev"
-Requires-Dist: ruff<1.0.0,>=0.12.0; extra == "dev"
-Requires-Dist: mypy<2.0.0,>=1.17.0; extra == "dev"
+Requires-Dist: pytest<10.0.0,>=9.2.0; extra == "dev"
+Requires-Dist: pytest-asyncio<2.0.0,>=1.5.0; extra == "dev"
+Requires-Dist: pytest-cov<8.0.0,>=7.1.0; extra == "dev"
+Requires-Dist: transformers<5.21.0,>=5.16.1; extra == "dev"
+Requires-Dist: black<27.0.0,>=26.10.1; extra == "dev"
+Requires-Dist: ruff<1.0.0,>=0.17.5; extra == "dev"
+Requires-Dist: mypy<3.0.0,>=2.4.1; extra == "dev"
 Provides-Extra: all
-Requires-Dist: vllm<0.32.0,>=0.31.0; extra == "all"
-Requires-Dist: torch==2.13.0; extra == "all"
-Requires-Dist: torchvision==0.28.0; extra == "all"
+Requires-Dist: vllm<0.34.0,>=0.33.0; extra == "all"
+Requires-Dist: torch==2.14.0; extra == "all"
+Requires-Dist: torchvision==0.29.0; extra == "all"
 Requires-Dist: torchaudio==2.11.0; extra == "all"
-Requires-Dist: transformers<5.18.0,>=5.10.4; extra == "all"
+Requires-Dist: transformers<5.21.0,>=5.16.1; extra == "all"
 Requires-Dist: qwen-vl-utils<0.1.0,>=0.0.14; extra == "all"
 Dynamic: author
 Dynamic: license-file
@@ -128,11 +129,18 @@ pip install --upgrade pip
 pip install "caption-flow[vllm]"
 ```
 
-The Linux GPU worker extra targets vLLM 0.31.x with PyTorch 2.13.0,
-torchvision 0.28.0, and torchaudio 2.11.0 (as required by vLLM), plus
-Transformers 5.10.4–5.17.x. To upgrade an existing worker from a checkout,
+The Linux GPU worker extra targets vLLM 0.33.x with PyTorch 2.14.0,
+torchvision 0.29.0, and torchaudio 2.11.0 (as required by vLLM), plus
+Transformers 5.16.1–5.20.x and Pillow 13. To upgrade an existing worker from a checkout,
 run `pip install --upgrade -e ".[vllm]"` in its environment and restart it.
 
+The tested worker stack uses NumPy 2.3.x on Python 3.12 and 2.4.x
+on Python 3.13, and FastAPI below 0.137. These bounds are shared with the
+base install so adding the worker extra remains compatible. CI resolves
+the full Linux worker and development dependencies on both Python versions.
+Lance 8.x, DuckDB 1.x, and Hugging Face Hub 1.x remain bounded to their
+existing major versions pending separate migration testing.
+
 An OpenAI-compatible API worker does not need the GPU dependencies:
 
 ```bash
@@ -144,26 +152,36 @@ For an orchestrator or monitor-only install, use `pip install -e .`.
 SimpleTuner. Terminal image previews remain optional because the current
 `term-image` release requires an older Pillow major than CaptionFlow uses.
 
-On Apple Silicon, install the MPS-compatible PyTorch chain and pinned Metal
-plugin with:
+On Apple Silicon with macOS 14 or newer, install the MPS-compatible PyTorch
+chain and image utilities with:
 
 ```bash
 pip install -e ".[apple]"
 ```
 
-CaptionFlow can use vLLM on Apple Silicon too, but the normal Linux `vllm`
-wheel is not the Apple install path. The pinned `vllm-metal` dependency is
-selected automatically on native arm64 Python 3.12+. For a ready-to-run
-Metal worker, use the upstream installer, which also builds/installs the
-Apple-specific vLLM core:
+For native vLLM Metal captioning, use **macOS 15+, Apple Silicon, and Python
+3.12**. The current Metal wheels are published on
+[GitHub releases](https://github.com/vllm-project/vllm-metal/releases), while
+PyPI still carries the obsolete 0.1.0 plugin. From this checkout, install the
+matching stable vLLM 0.33.0 core and Metal 0.33.0 wheels into your environment:
 
 ```bash
-curl -fsSL https://raw.githubusercontent.com/vllm-project/vllm-metal/main/install.sh | bash
-source ~/.venv-vllm-metal/bin/activate
-pip install -e .
+python -m pip install --upgrade -r requirements.metal.txt
+python -m pip install -e ".[dev]"
+python -m pip check
+python examples/metal_smoke.py
 ```
 
-Do not combine `.[apple]` with the Linux `.[vllm]` extra.
+The smoke test downloads `mlx-community/Qwen3-VL-2B-Instruct-4bit` on first
+use and runs two images through CaptionFlow's native caption pipeline. It
+checks that the Metal plugin and MLX GPU are active, and that the captions
+identify the images' shapes and different colors. Pass `--model /path/to/model`
+to use a local checkpoint. The test does not connect to an orchestrator.
+
+The stable recipe uses the core wheel's PyTorch 2.14.0 pin. Upstream's installer
+can select newer development builds and override Torch with nightly packages;
+those are separate from this recipe. `.[apple]` alone does not install a vLLM
+runtime, and should not be combined with the Linux `.[vllm]` extra.
 
 For native macOS CPU vLLM instead, follow the
 [official source-build instructions](https://docs.vllm.ai/en/stable/getting_started/installation/cpu/?device=apple).
@@ -225,7 +243,7 @@ Options:
 * **csv**: exports CSV-compatible data columns to the `--output` path containing incomplete metadata
 * **json**: creates a `.json` file for each sample inside the `--output` subdirectory containing **complete** metadata; useful for webdatasets
 * **txt**: creates `.txt` file for each sample inside the `--output` subdirectory containing ONLY captions
-* **webshart**: updates an **existing per-shard metadata `.json` file** by writing captions under the plural `captions` key. Strings, native JSON objects, and mixed lists are supported. JSON objects stored as text in Lance output columns are decoded during export, preserving nested values instead of writing Python dictionary representations. Native objects supplied through `Caption.outputs` are encoded as JSON only for Lance persistence. Requires webshart 0.6.0 or newer for native JSON caption support (included in the base dependencies). For this format, pass `--output` as the path to the existing shard metadata JSON file when exporting one shard. If you export multiple shards, pass `--output` as a directory containing one existing `{shard_name}.json` file per shard.
+* **webshart**: updates an **existing per-shard metadata `.json` file** by writing captions under the plural `captions` key. Strings, native JSON objects, and mixed lists are supported. JSON objects stored as text in Lance output columns are decoded during export, preserving nested values instead of writing Python dictionary representations. Native objects supplied through `Caption.outputs` are encoded as JSON only for Lance persistence. Requires webshart 0.7.x (included in the base dependencies), including native JSON caption support. For this format, pass `--output` as the path to the existing shard metadata JSON file when exporting one shard. If you export multiple shards, pass `--output` as a directory containing one existing `{shard_name}.json` file per shard.
 * **huggingface_hub**: creates a dataset on Hugging Face Hub, possibly `--private` and `--nsfw` where necessary
 * **all**: creates the directory/file-generating export formats in a specified `--output` directory. prefer a directory here; `webshart` is a special case that expects existing per-shard metadata `.json` files rather than creating new metadata files.
 

{caption_flow-0.5.7 → caption_flow-0.6.1}/README.md

@@ -27,11 +27,18 @@ pip install --upgrade pip
 pip install "caption-flow[vllm]"
 ```
 
-The Linux GPU worker extra targets vLLM 0.31.x with PyTorch 2.13.0,
-torchvision 0.28.0, and torchaudio 2.11.0 (as required by vLLM), plus
-Transformers 5.10.4–5.17.x. To upgrade an existing worker from a checkout,
+The Linux GPU worker extra targets vLLM 0.33.x with PyTorch 2.14.0,
+torchvision 0.29.0, and torchaudio 2.11.0 (as required by vLLM), plus
+Transformers 5.16.1–5.20.x and Pillow 13. To upgrade an existing worker from a checkout,
 run `pip install --upgrade -e ".[vllm]"` in its environment and restart it.
 
+The tested worker stack uses NumPy 2.3.x on Python 3.12 and 2.4.x
+on Python 3.13, and FastAPI below 0.137. These bounds are shared with the
+base install so adding the worker extra remains compatible. CI resolves
+the full Linux worker and development dependencies on both Python versions.
+Lance 8.x, DuckDB 1.x, and Hugging Face Hub 1.x remain bounded to their
+existing major versions pending separate migration testing.
+
 An OpenAI-compatible API worker does not need the GPU dependencies:
 
 ```bash
@@ -43,26 +50,36 @@ For an orchestrator or monitor-only install, use `pip install -e .`.
 SimpleTuner. Terminal image previews remain optional because the current
 `term-image` release requires an older Pillow major than CaptionFlow uses.
 
-On Apple Silicon, install the MPS-compatible PyTorch chain and pinned Metal
-plugin with:
+On Apple Silicon with macOS 14 or newer, install the MPS-compatible PyTorch
+chain and image utilities with:
 
 ```bash
 pip install -e ".[apple]"
 ```
 
-CaptionFlow can use vLLM on Apple Silicon too, but the normal Linux `vllm`
-wheel is not the Apple install path. The pinned `vllm-metal` dependency is
-selected automatically on native arm64 Python 3.12+. For a ready-to-run
-Metal worker, use the upstream installer, which also builds/installs the
-Apple-specific vLLM core:
+For native vLLM Metal captioning, use **macOS 15+, Apple Silicon, and Python
+3.12**. The current Metal wheels are published on
+[GitHub releases](https://github.com/vllm-project/vllm-metal/releases), while
+PyPI still carries the obsolete 0.1.0 plugin. From this checkout, install the
+matching stable vLLM 0.33.0 core and Metal 0.33.0 wheels into your environment:
 
 ```bash
-curl -fsSL https://raw.githubusercontent.com/vllm-project/vllm-metal/main/install.sh | bash
-source ~/.venv-vllm-metal/bin/activate
-pip install -e .
+python -m pip install --upgrade -r requirements.metal.txt
+python -m pip install -e ".[dev]"
+python -m pip check
+python examples/metal_smoke.py
 ```
 
-Do not combine `.[apple]` with the Linux `.[vllm]` extra.
+The smoke test downloads `mlx-community/Qwen3-VL-2B-Instruct-4bit` on first
+use and runs two images through CaptionFlow's native caption pipeline. It
+checks that the Metal plugin and MLX GPU are active, and that the captions
+identify the images' shapes and different colors. Pass `--model /path/to/model`
+to use a local checkpoint. The test does not connect to an orchestrator.
+
+The stable recipe uses the core wheel's PyTorch 2.14.0 pin. Upstream's installer
+can select newer development builds and override Torch with nightly packages;
+those are separate from this recipe. `.[apple]` alone does not install a vLLM
+runtime, and should not be combined with the Linux `.[vllm]` extra.
 
 For native macOS CPU vLLM instead, follow the
 [official source-build instructions](https://docs.vllm.ai/en/stable/getting_started/installation/cpu/?device=apple).
@@ -124,7 +141,7 @@ Options:
 * **csv**: exports CSV-compatible data columns to the `--output` path containing incomplete metadata
 * **json**: creates a `.json` file for each sample inside the `--output` subdirectory containing **complete** metadata; useful for webdatasets
 * **txt**: creates `.txt` file for each sample inside the `--output` subdirectory containing ONLY captions
-* **webshart**: updates an **existing per-shard metadata `.json` file** by writing captions under the plural `captions` key. Strings, native JSON objects, and mixed lists are supported. JSON objects stored as text in Lance output columns are decoded during export, preserving nested values instead of writing Python dictionary representations. Native objects supplied through `Caption.outputs` are encoded as JSON only for Lance persistence. Requires webshart 0.6.0 or newer for native JSON caption support (included in the base dependencies). For this format, pass `--output` as the path to the existing shard metadata JSON file when exporting one shard. If you export multiple shards, pass `--output` as a directory containing one existing `{shard_name}.json` file per shard.
+* **webshart**: updates an **existing per-shard metadata `.json` file** by writing captions under the plural `captions` key. Strings, native JSON objects, and mixed lists are supported. JSON objects stored as text in Lance output columns are decoded during export, preserving nested values instead of writing Python dictionary representations. Native objects supplied through `Caption.outputs` are encoded as JSON only for Lance persistence. Requires webshart 0.7.x (included in the base dependencies), including native JSON caption support. For this format, pass `--output` as the path to the existing shard metadata JSON file when exporting one shard. If you export multiple shards, pass `--output` as a directory containing one existing `{shard_name}.json` file per shard.
 * **huggingface_hub**: creates a dataset on Hugging Face Hub, possibly `--private` and `--nsfw` where necessary
 * **all**: creates the directory/file-generating export formats in a specified `--output` directory. prefer a directory here; `webshart` is a special case that expects existing per-shard metadata `.json` files rather than creating new metadata files.
 

{caption_flow-0.5.7 → caption_flow-0.6.1}/pyproject.toml

@@ -1,5 +1,5 @@
 [build-system]
-requires = ["setuptools>=68.0.0", "wheel"]
+requires = ["setuptools>=77.0.3", "wheel"]
 build-backend = "setuptools.build_meta"
 
 [project]

{caption_flow-0.5.7 → caption_flow-0.6.1}/setup.py

@@ -23,58 +23,61 @@ def get_version() -> str:
 
 
 base_deps = [
-    "websockets>=16.0,<17.0",
-    "pyarrow>=21.0.0,<26.0.0",
-    "click>=8.2.0,<9.0.0",
-    "pydantic>=2.12.0,<3.0.0",
-    "aiofiles>=24.1.0,<26.0.0",
-    "rich>=14.0.0,<16.0.0",
-    "cryptography>=45.0.0,<50.0.0",
-    "PyYAML>=6.0.2,<7.0.0",
-    "certbot>=5.0.0,<6.0.0",
-    "numpy>=2.2.0,<3.0.0",
-    "Pillow>=11.3.0,<13.0.0",
-    "trainingsample>=0.3.2,<0.4.0",
-    "pandas>=3.0.0,<4.0.0",
-    "datasets>=5.0.0,<6.0.0",
-    "boto3>=1.43.0,<2.0.0",
-    "webshart>=0.6.0,<0.7.0",
-    "pylance>=8.0.0,<9.0.0",
-    "duckdb>=1.5.0,<2.0.0",
-    "aiohttp>=3.13.3,<4.0.0",
-    "fastapi>=0.133.0,<0.137.0",
-    "uvicorn>=0.35.0,<1.0.0",
-    "huggingface-hub>=1.5.0,<2.0.0",
-    "psutil>=7.0.0,<8.0.0",
-    "requests>=2.32.0,<3.0.0",
-    "tqdm>=4.67.0,<5.0.0",
-    "urwid>=3.0.2,<5.0.0",
+    "websockets>=17.4,<18.0",
+    "pyarrow>=26.0.1,<27.0.0",
+    "click>=8.5.1,<9.0.0",
+    "pydantic>=2.14.3,<3.0.0",
+    "aiofiles>=25.1.0,<26.0.0",
+    "rich>=15.0.0,<16.0.0",
+    "cryptography>=50.0.2,<51.0.0",
+    "PyYAML>=6.0.3,<7.0.0",
+    "certbot>=5.10.0,<6.0.0",
+    # vLLM's mistral-common requires NumPy <2.4 on Python 3.12;
+    # its pinned numba 0.65 requires NumPy <2.5 on Python 3.13.
+    "numpy>=2.3.5,<2.4.0; python_version < '3.13'",
+    "numpy>=2.4.6,<2.5.0; python_version >= '3.13'",
+    "Pillow>=13.0.0,<14.0.0",
+    "trainingsample==0.4.0",
+    "pandas>=3.1.1,<4.0.0",
+    "datasets>=5.2.0,<6.0.0",
+    "boto3>=1.44.3,<2.0.0",
+    "webshart>=0.7.0,<0.8.0",
+    "pylance>=8.0.1,<9.0.0",
+    "duckdb>=1.5.6,<2.0.0",
+    "aiohttp>=3.14.5,<4.0.0",
+    # vLLM 0.33 requires FastAPI below 0.137.
+    "fastapi>=0.136.3,<0.137.0",
+    "uvicorn>=0.54.0,<1.0.0",
+    "huggingface-hub>=1.33.0,<2.0.0",
+    "psutil>=7.2.2,<8.0.0",
+    "requests>=2.34.2,<3.0.0",
+    "tqdm>=4.70.1,<5.0.0",
+    "urwid>=5.1.3,<6.0.0",
 ]
 
-# Match vLLM 0.31's published requirements, including torchaudio 2.11.0.
+# Match vLLM 0.33's published requirements, including torchaudio 2.11.0.
 # Keep these explicit so a worker install cannot silently combine incompatible
 # torch packages when another dependency broadens its requirements.
 vllm_deps = [
-    "vllm>=0.31.0,<0.32.0",
-    "torch==2.13.0",
-    "torchvision==0.28.0",
+    "vllm>=0.33.0,<0.34.0",
+    "torch==2.14.0",
+    "torchvision==0.29.0",
     "torchaudio==2.11.0",
-    "transformers>=5.10.4,<5.18.0",
+    "transformers>=5.16.1,<5.21.0",
     "qwen-vl-utils>=0.0.14,<0.1.0",
 ]
 
 PYTORCH_DEPENDENCIES = [
-    "torch>=2.13.0",
-    "torchvision>=0.28.0",
+    "torch>=2.14.0",
+    "torchvision>=0.29.0",
     "torchaudio>=2.11.0",
 ]
 
 APPLE_PYTORCH_DEPENDENCIES = [
-    "torch>=2.13.0",
-    "torchvision>=0.28.0",
-    "torchaudio>=2.11.0",
-    # vLLM Metal publishes Apple Silicon wheels for Python 3.12+.
-    "vllm-metal==0.1.0; platform_system == 'Darwin' and platform_machine == 'arm64' and python_version >= '3.12'",
+    *PYTORCH_DEPENDENCIES,
+    "qwen-vl-utils>=0.0.14,<0.1.0",
+    # Install the matching vLLM core and Metal wheels from upstream GitHub
+    # releases separately; PyPI's vllm-metal 0.1.0 is obsolete.
 ]
 
 extras_require = {
@@ -89,13 +92,13 @@ extras_require = {
     "rocm": PYTORCH_DEPENDENCIES,
     "apple": APPLE_PYTORCH_DEPENDENCIES,
     "dev": [
-        "pytest>=8.0.0,<9.0.0",
-        "pytest-asyncio>=1.1.0,<2.0.0",
-        "pytest-cov>=6.0.0,<7.0.0",
-        "transformers>=5.10.4,<5.18.0",
-        "black>=25.0.0,<26.0.0",
-        "ruff>=0.12.0,<1.0.0",
-        "mypy>=1.17.0,<2.0.0",
+        "pytest>=9.2.0,<10.0.0",
+        "pytest-asyncio>=1.5.0,<2.0.0",
+        "pytest-cov>=7.1.0,<8.0.0",
+        "transformers>=5.16.1,<5.21.0",
+        "black>=26.10.1,<27.0.0",
+        "ruff>=0.17.5,<1.0.0",
+        "mypy>=2.4.1,<3.0.0",
     ],
     "all": [*vllm_deps],
 }

{caption_flow-0.5.7 → caption_flow-0.6.1}/src/caption_flow/__init__.py

@@ -1,6 +1,6 @@
 """CaptionFlow - Distributed community captioning system."""
 
-__version__ = "0.5.7"
+__version__ = "0.6.1"
 
 from .monitor import Monitor
 from .orchestrator import Orchestrator

{caption_flow-0.5.7 → caption_flow-0.6.1}/src/caption_flow.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: caption-flow
-Version: 0.5.7
+Version: 0.6.1
 Summary: Self-contained distributed community captioning system
 Author: bghira
 Author-email: bghira <bghira@users.github.com>
@@ -16,82 +16,83 @@ Classifier: Programming Language :: Python :: 3.13
 Requires-Python: >=3.12,<3.14
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: websockets<17.0,>=16.0
-Requires-Dist: pyarrow<26.0.0,>=21.0.0
-Requires-Dist: click<9.0.0,>=8.2.0
-Requires-Dist: pydantic<3.0.0,>=2.12.0
-Requires-Dist: aiofiles<26.0.0,>=24.1.0
-Requires-Dist: rich<16.0.0,>=14.0.0
-Requires-Dist: cryptography<50.0.0,>=45.0.0
-Requires-Dist: PyYAML<7.0.0,>=6.0.2
-Requires-Dist: certbot<6.0.0,>=5.0.0
-Requires-Dist: numpy<3.0.0,>=2.2.0
-Requires-Dist: Pillow<13.0.0,>=11.3.0
-Requires-Dist: trainingsample<0.4.0,>=0.3.2
-Requires-Dist: pandas<4.0.0,>=3.0.0
-Requires-Dist: datasets<6.0.0,>=5.0.0
-Requires-Dist: boto3<2.0.0,>=1.43.0
-Requires-Dist: webshart<0.7.0,>=0.6.0
-Requires-Dist: pylance<9.0.0,>=8.0.0
-Requires-Dist: duckdb<2.0.0,>=1.5.0
-Requires-Dist: aiohttp<4.0.0,>=3.13.3
-Requires-Dist: fastapi<0.137.0,>=0.133.0
-Requires-Dist: uvicorn<1.0.0,>=0.35.0
-Requires-Dist: huggingface-hub<2.0.0,>=1.5.0
-Requires-Dist: psutil<8.0.0,>=7.0.0
-Requires-Dist: requests<3.0.0,>=2.32.0
-Requires-Dist: tqdm<5.0.0,>=4.67.0
-Requires-Dist: urwid<5.0.0,>=3.0.2
+Requires-Dist: websockets<18.0,>=17.4
+Requires-Dist: pyarrow<27.0.0,>=26.0.1
+Requires-Dist: click<9.0.0,>=8.5.1
+Requires-Dist: pydantic<3.0.0,>=2.14.3
+Requires-Dist: aiofiles<26.0.0,>=25.1.0
+Requires-Dist: rich<16.0.0,>=15.0.0
+Requires-Dist: cryptography<51.0.0,>=50.0.2
+Requires-Dist: PyYAML<7.0.0,>=6.0.3
+Requires-Dist: certbot<6.0.0,>=5.10.0
+Requires-Dist: numpy<2.4.0,>=2.3.5; python_version < "3.13"
+Requires-Dist: numpy<2.5.0,>=2.4.6; python_version >= "3.13"
+Requires-Dist: Pillow<14.0.0,>=13.0.0
+Requires-Dist: trainingsample==0.4.0
+Requires-Dist: pandas<4.0.0,>=3.1.1
+Requires-Dist: datasets<6.0.0,>=5.2.0
+Requires-Dist: boto3<2.0.0,>=1.44.3
+Requires-Dist: webshart<0.8.0,>=0.7.0
+Requires-Dist: pylance<9.0.0,>=8.0.1
+Requires-Dist: duckdb<2.0.0,>=1.5.6
+Requires-Dist: aiohttp<4.0.0,>=3.14.5
+Requires-Dist: fastapi<0.137.0,>=0.136.3
+Requires-Dist: uvicorn<1.0.0,>=0.54.0
+Requires-Dist: huggingface-hub<2.0.0,>=1.33.0
+Requires-Dist: psutil<8.0.0,>=7.2.2
+Requires-Dist: requests<3.0.0,>=2.34.2
+Requires-Dist: tqdm<5.0.0,>=4.70.1
+Requires-Dist: urwid<6.0.0,>=5.1.3
 Provides-Extra: vllm
-Requires-Dist: vllm<0.32.0,>=0.31.0; extra == "vllm"
-Requires-Dist: torch==2.13.0; extra == "vllm"
-Requires-Dist: torchvision==0.28.0; extra == "vllm"
+Requires-Dist: vllm<0.34.0,>=0.33.0; extra == "vllm"
+Requires-Dist: torch==2.14.0; extra == "vllm"
+Requires-Dist: torchvision==0.29.0; extra == "vllm"
 Requires-Dist: torchaudio==2.11.0; extra == "vllm"
-Requires-Dist: transformers<5.18.0,>=5.10.4; extra == "vllm"
+Requires-Dist: transformers<5.21.0,>=5.16.1; extra == "vllm"
 Requires-Dist: qwen-vl-utils<0.1.0,>=0.0.14; extra == "vllm"
 Provides-Extra: captioning
-Requires-Dist: vllm<0.32.0,>=0.31.0; extra == "captioning"
-Requires-Dist: torch==2.13.0; extra == "captioning"
-Requires-Dist: torchvision==0.28.0; extra == "captioning"
+Requires-Dist: vllm<0.34.0,>=0.33.0; extra == "captioning"
+Requires-Dist: torch==2.14.0; extra == "captioning"
+Requires-Dist: torchvision==0.29.0; extra == "captioning"
 Requires-Dist: torchaudio==2.11.0; extra == "captioning"
-Requires-Dist: transformers<5.18.0,>=5.10.4; extra == "captioning"
+Requires-Dist: transformers<5.21.0,>=5.16.1; extra == "captioning"
 Requires-Dist: qwen-vl-utils<0.1.0,>=0.0.14; extra == "captioning"
 Provides-Extra: openai
 Provides-Extra: cpu
-Requires-Dist: torch>=2.13.0; extra == "cpu"
-Requires-Dist: torchvision>=0.28.0; extra == "cpu"
+Requires-Dist: torch>=2.14.0; extra == "cpu"
+Requires-Dist: torchvision>=0.29.0; extra == "cpu"
 Requires-Dist: torchaudio>=2.11.0; extra == "cpu"
 Provides-Extra: cuda
-Requires-Dist: torch>=2.13.0; extra == "cuda"
-Requires-Dist: torchvision>=0.28.0; extra == "cuda"
+Requires-Dist: torch>=2.14.0; extra == "cuda"
+Requires-Dist: torchvision>=0.29.0; extra == "cuda"
 Requires-Dist: torchaudio>=2.11.0; extra == "cuda"
 Provides-Extra: cuda13
-Requires-Dist: torch>=2.13.0; extra == "cuda13"
-Requires-Dist: torchvision>=0.28.0; extra == "cuda13"
+Requires-Dist: torch>=2.14.0; extra == "cuda13"
+Requires-Dist: torchvision>=0.29.0; extra == "cuda13"
 Requires-Dist: torchaudio>=2.11.0; extra == "cuda13"
 Provides-Extra: rocm
-Requires-Dist: torch>=2.13.0; extra == "rocm"
-Requires-Dist: torchvision>=0.28.0; extra == "rocm"
+Requires-Dist: torch>=2.14.0; extra == "rocm"
+Requires-Dist: torchvision>=0.29.0; extra == "rocm"
 Requires-Dist: torchaudio>=2.11.0; extra == "rocm"
 Provides-Extra: apple
-Requires-Dist: torch>=2.13.0; extra == "apple"
-Requires-Dist: torchvision>=0.28.0; extra == "apple"
+Requires-Dist: torch>=2.14.0; extra == "apple"
+Requires-Dist: torchvision>=0.29.0; extra == "apple"
 Requires-Dist: torchaudio>=2.11.0; extra == "apple"
-Requires-Dist: vllm-metal==0.1.0; (platform_system == "Darwin" and platform_machine == "arm64" and python_version >= "3.12") and extra == "apple"
+Requires-Dist: qwen-vl-utils<0.1.0,>=0.0.14; extra == "apple"
 Provides-Extra: dev
-Requires-Dist: pytest<9.0.0,>=8.0.0; extra == "dev"
-Requires-Dist: pytest-asyncio<2.0.0,>=1.1.0; extra == "dev"
-Requires-Dist: pytest-cov<7.0.0,>=6.0.0; extra == "dev"
-Requires-Dist: transformers<5.18.0,>=5.10.4; extra == "dev"
-Requires-Dist: black<26.0.0,>=25.0.0; extra == "dev"
-Requires-Dist: ruff<1.0.0,>=0.12.0; extra == "dev"
-Requires-Dist: mypy<2.0.0,>=1.17.0; extra == "dev"
+Requires-Dist: pytest<10.0.0,>=9.2.0; extra == "dev"
+Requires-Dist: pytest-asyncio<2.0.0,>=1.5.0; extra == "dev"
+Requires-Dist: pytest-cov<8.0.0,>=7.1.0; extra == "dev"
+Requires-Dist: transformers<5.21.0,>=5.16.1; extra == "dev"
+Requires-Dist: black<27.0.0,>=26.10.1; extra == "dev"
+Requires-Dist: ruff<1.0.0,>=0.17.5; extra == "dev"
+Requires-Dist: mypy<3.0.0,>=2.4.1; extra == "dev"
 Provides-Extra: all
-Requires-Dist: vllm<0.32.0,>=0.31.0; extra == "all"
-Requires-Dist: torch==2.13.0; extra == "all"
-Requires-Dist: torchvision==0.28.0; extra == "all"
+Requires-Dist: vllm<0.34.0,>=0.33.0; extra == "all"
+Requires-Dist: torch==2.14.0; extra == "all"
+Requires-Dist: torchvision==0.29.0; extra == "all"
 Requires-Dist: torchaudio==2.11.0; extra == "all"
-Requires-Dist: transformers<5.18.0,>=5.10.4; extra == "all"
+Requires-Dist: transformers<5.21.0,>=5.16.1; extra == "all"
 Requires-Dist: qwen-vl-utils<0.1.0,>=0.0.14; extra == "all"
 Dynamic: author
 Dynamic: license-file
@@ -128,11 +129,18 @@ pip install --upgrade pip
 pip install "caption-flow[vllm]"
 ```
 
-The Linux GPU worker extra targets vLLM 0.31.x with PyTorch 2.13.0,
-torchvision 0.28.0, and torchaudio 2.11.0 (as required by vLLM), plus
-Transformers 5.10.4–5.17.x. To upgrade an existing worker from a checkout,
+The Linux GPU worker extra targets vLLM 0.33.x with PyTorch 2.14.0,
+torchvision 0.29.0, and torchaudio 2.11.0 (as required by vLLM), plus
+Transformers 5.16.1–5.20.x and Pillow 13. To upgrade an existing worker from a checkout,
 run `pip install --upgrade -e ".[vllm]"` in its environment and restart it.
 
+The tested worker stack uses NumPy 2.3.x on Python 3.12 and 2.4.x
+on Python 3.13, and FastAPI below 0.137. These bounds are shared with the
+base install so adding the worker extra remains compatible. CI resolves
+the full Linux worker and development dependencies on both Python versions.
+Lance 8.x, DuckDB 1.x, and Hugging Face Hub 1.x remain bounded to their
+existing major versions pending separate migration testing.
+
 An OpenAI-compatible API worker does not need the GPU dependencies:
 
 ```bash
@@ -144,26 +152,36 @@ For an orchestrator or monitor-only install, use `pip install -e .`.
 SimpleTuner. Terminal image previews remain optional because the current
 `term-image` release requires an older Pillow major than CaptionFlow uses.
 
-On Apple Silicon, install the MPS-compatible PyTorch chain and pinned Metal
-plugin with:
+On Apple Silicon with macOS 14 or newer, install the MPS-compatible PyTorch
+chain and image utilities with:
 
 ```bash
 pip install -e ".[apple]"
 ```
 
-CaptionFlow can use vLLM on Apple Silicon too, but the normal Linux `vllm`
-wheel is not the Apple install path. The pinned `vllm-metal` dependency is
-selected automatically on native arm64 Python 3.12+. For a ready-to-run
-Metal worker, use the upstream installer, which also builds/installs the
-Apple-specific vLLM core:
+For native vLLM Metal captioning, use **macOS 15+, Apple Silicon, and Python
+3.12**. The current Metal wheels are published on
+[GitHub releases](https://github.com/vllm-project/vllm-metal/releases), while
+PyPI still carries the obsolete 0.1.0 plugin. From this checkout, install the
+matching stable vLLM 0.33.0 core and Metal 0.33.0 wheels into your environment:
 
 ```bash
-curl -fsSL https://raw.githubusercontent.com/vllm-project/vllm-metal/main/install.sh | bash
-source ~/.venv-vllm-metal/bin/activate
-pip install -e .
+python -m pip install --upgrade -r requirements.metal.txt
+python -m pip install -e ".[dev]"
+python -m pip check
+python examples/metal_smoke.py
 ```
 
-Do not combine `.[apple]` with the Linux `.[vllm]` extra.
+The smoke test downloads `mlx-community/Qwen3-VL-2B-Instruct-4bit` on first
+use and runs two images through CaptionFlow's native caption pipeline. It
+checks that the Metal plugin and MLX GPU are active, and that the captions
+identify the images' shapes and different colors. Pass `--model /path/to/model`
+to use a local checkpoint. The test does not connect to an orchestrator.
+
+The stable recipe uses the core wheel's PyTorch 2.14.0 pin. Upstream's installer
+can select newer development builds and override Torch with nightly packages;
+those are separate from this recipe. `.[apple]` alone does not install a vLLM
+runtime, and should not be combined with the Linux `.[vllm]` extra.
 
 For native macOS CPU vLLM instead, follow the
 [official source-build instructions](https://docs.vllm.ai/en/stable/getting_started/installation/cpu/?device=apple).
@@ -225,7 +243,7 @@ Options:
 * **csv**: exports CSV-compatible data columns to the `--output` path containing incomplete metadata
 * **json**: creates a `.json` file for each sample inside the `--output` subdirectory containing **complete** metadata; useful for webdatasets
 * **txt**: creates `.txt` file for each sample inside the `--output` subdirectory containing ONLY captions
-* **webshart**: updates an **existing per-shard metadata `.json` file** by writing captions under the plural `captions` key. Strings, native JSON objects, and mixed lists are supported. JSON objects stored as text in Lance output columns are decoded during export, preserving nested values instead of writing Python dictionary representations. Native objects supplied through `Caption.outputs` are encoded as JSON only for Lance persistence. Requires webshart 0.6.0 or newer for native JSON caption support (included in the base dependencies). For this format, pass `--output` as the path to the existing shard metadata JSON file when exporting one shard. If you export multiple shards, pass `--output` as a directory containing one existing `{shard_name}.json` file per shard.
+* **webshart**: updates an **existing per-shard metadata `.json` file** by writing captions under the plural `captions` key. Strings, native JSON objects, and mixed lists are supported. JSON objects stored as text in Lance output columns are decoded during export, preserving nested values instead of writing Python dictionary representations. Native objects supplied through `Caption.outputs` are encoded as JSON only for Lance persistence. Requires webshart 0.7.x (included in the base dependencies), including native JSON caption support. For this format, pass `--output` as the path to the existing shard metadata JSON file when exporting one shard. If you export multiple shards, pass `--output` as a directory containing one existing `{shard_name}.json` file per shard.
 * **huggingface_hub**: creates a dataset on Hugging Face Hub, possibly `--private` and `--nsfw` where necessary
 * **all**: creates the directory/file-generating export formats in a specified `--output` directory. prefer a directory here; `webshart` is a special case that expects existing per-shard metadata `.json` files rather than creating new metadata files.
 

caption_flow-0.6.1/src/caption_flow.egg-info/requires.txt

@@ -0,0 +1,92 @@
+websockets<18.0,>=17.4
+pyarrow<27.0.0,>=26.0.1
+click<9.0.0,>=8.5.1
+pydantic<3.0.0,>=2.14.3
+aiofiles<26.0.0,>=25.1.0
+rich<16.0.0,>=15.0.0
+cryptography<51.0.0,>=50.0.2
+PyYAML<7.0.0,>=6.0.3
+certbot<6.0.0,>=5.10.0
+Pillow<14.0.0,>=13.0.0
+trainingsample==0.4.0
+pandas<4.0.0,>=3.1.1
+datasets<6.0.0,>=5.2.0
+boto3<2.0.0,>=1.44.3
+webshart<0.8.0,>=0.7.0
+pylance<9.0.0,>=8.0.1
+duckdb<2.0.0,>=1.5.6
+aiohttp<4.0.0,>=3.14.5
+fastapi<0.137.0,>=0.136.3
+uvicorn<1.0.0,>=0.54.0
+huggingface-hub<2.0.0,>=1.33.0
+psutil<8.0.0,>=7.2.2
+requests<3.0.0,>=2.34.2
+tqdm<5.0.0,>=4.70.1
+urwid<6.0.0,>=5.1.3
+
+[:python_version < "3.13"]
+numpy<2.4.0,>=2.3.5
+
+[:python_version >= "3.13"]
+numpy<2.5.0,>=2.4.6
+
+[all]
+vllm<0.34.0,>=0.33.0
+torch==2.14.0
+torchvision==0.29.0
+torchaudio==2.11.0
+transformers<5.21.0,>=5.16.1
+qwen-vl-utils<0.1.0,>=0.0.14
+
+[apple]
+torch>=2.14.0
+torchvision>=0.29.0
+torchaudio>=2.11.0
+qwen-vl-utils<0.1.0,>=0.0.14
+
+[captioning]
+vllm<0.34.0,>=0.33.0
+torch==2.14.0
+torchvision==0.29.0
+torchaudio==2.11.0
+transformers<5.21.0,>=5.16.1
+qwen-vl-utils<0.1.0,>=0.0.14
+
+[cpu]
+torch>=2.14.0
+torchvision>=0.29.0
+torchaudio>=2.11.0
+
+[cuda]
+torch>=2.14.0
+torchvision>=0.29.0
+torchaudio>=2.11.0
+
+[cuda13]
+torch>=2.14.0
+torchvision>=0.29.0
+torchaudio>=2.11.0
+
+[dev]
+pytest<10.0.0,>=9.2.0
+pytest-asyncio<2.0.0,>=1.5.0
+pytest-cov<8.0.0,>=7.1.0
+transformers<5.21.0,>=5.16.1
+black<27.0.0,>=26.10.1
+ruff<1.0.0,>=0.17.5
+mypy<3.0.0,>=2.4.1
+
+[openai]
+
+[rocm]
+torch>=2.14.0
+torchvision>=0.29.0
+torchaudio>=2.11.0
+
+[vllm]
+vllm<0.34.0,>=0.33.0
+torch==2.14.0
+torchvision==0.29.0
+torchaudio==2.11.0
+transformers<5.21.0,>=5.16.1
+qwen-vl-utils<0.1.0,>=0.0.14

{caption_flow-0.5.7 → caption_flow-0.6.1}/tests/test_dependency_config.py

@@ -27,12 +27,12 @@ def requirements_by_name(requirements):
 @pytest.mark.parametrize("extra", ["vllm", "captioning", "all"])
 def test_vllm_extras_select_matching_runtime(package_config, extra):
     dependencies = requirements_by_name(package_config["extras_require"][extra])
-    assert dependencies["vllm"].specifier.contains("0.31.0")
-    assert not dependencies["vllm"].specifier.contains("0.30.0")
+    assert dependencies["vllm"].specifier.contains("0.33.0")
     assert not dependencies["vllm"].specifier.contains("0.32.0")
+    assert not dependencies["vllm"].specifier.contains("0.34.0")
     for name, version in (
-        ("torch", "2.13.0"),
-        ("torchvision", "0.28.0"),
+        ("torch", "2.14.0"),
+        ("torchvision", "0.29.0"),
         ("torchaudio", "2.11.0"),
     ):
         assert str(dependencies[name].specifier) == f"=={version}"
@@ -42,19 +42,19 @@ def test_vllm_extras_select_matching_runtime(package_config, extra):
 def test_transformers_matches_vllm_supported_range(package_config, extra):
     dependencies = requirements_by_name(package_config["extras_require"][extra])
     versions = dependencies["transformers"].specifier
-    assert versions.contains("5.10.4")
-    assert versions.contains("5.17.0")
-    assert not versions.contains("5.10.3")
-    assert not versions.contains("5.18.0")
+    assert versions.contains("5.16.1")
+    assert versions.contains("5.20.2")
+    assert not versions.contains("5.16.0")
+    assert not versions.contains("5.21.0")
 
 
 @pytest.mark.parametrize("extra", ["cpu", "cuda", "cuda13", "rocm", "apple"])
 def test_pytorch_extras_require_updated_family(package_config, extra):
     dependencies = requirements_by_name(package_config["extras_require"][extra])
-    assert dependencies["torch"].specifier.contains("2.13.0")
-    assert not dependencies["torch"].specifier.contains("2.12.0")
-    assert dependencies["torchvision"].specifier.contains("0.28.0")
-    assert not dependencies["torchvision"].specifier.contains("0.27.0")
+    assert dependencies["torch"].specifier.contains("2.14.0")
+    assert not dependencies["torch"].specifier.contains("2.13.0")
+    assert dependencies["torchvision"].specifier.contains("0.29.0")
+    assert not dependencies["torchvision"].specifier.contains("0.28.0")
     assert dependencies["torchaudio"].specifier.contains("2.11.0")
     assert "vllm" not in dependencies
 

{caption_flow-0.5.7 → caption_flow-0.6.1}/tests/test_webshart_dependency.py

@@ -22,6 +22,7 @@ def test_base_dependency_requires_structured_caption_support():
         for dependency in dependencies
         if (requirement := Requirement(dependency)).name == "webshart"
     )
-    assert "0.6.0" in requirement.specifier
+    assert "0.7.0" in requirement.specifier
     assert "0.5.4" not in requirement.specifier
-    assert "0.7.0" not in requirement.specifier
+    assert "0.6.0" not in requirement.specifier
+    assert "0.8.0" not in requirement.specifier

{caption_flow-0.5.7 → caption_flow-0.6.1}/tests/test_worker_caption.py

@@ -508,7 +508,10 @@ class TestCaptionWorker:
         # Mock tokenizer and processor
         mock_tokenizer = Mock()
         mock_tokenizer.encode = Mock(return_value=[1] * 100)  # 100 tokens
+        # qwen-vl-utils uses the callable tokenizer; the fallback uses encode().
+        mock_tokenizer.return_value.input_ids = [1] * 100
         mock_processor = Mock()
+        mock_processor.apply_chat_template.return_value = "Describe this image"
         mock_sampling = Mock()
 
         stage = caption_worker.stages[0]
@@ -522,6 +525,13 @@ class TestCaptionWorker:
         assert len(processable) == 3
         assert len(too_long) == 0
 
+        # The 50-token safety buffer must still exclude these 100-token prompts.
+        processable, too_long = caption_worker._validate_and_split_batch(
+            batch, stage, mock_processor, mock_tokenizer, mock_sampling, max_length=150
+        )
+        assert processable == []
+        assert too_long == batch
+
     def test_resize_image_for_tokens(self, caption_worker):
         """Test image resizing for token reduction."""
         item = ProcessingItem(

caption_flow-0.5.7/src/caption_flow.egg-info/requires.txt

@@ -1,89 +0,0 @@
-websockets<17.0,>=16.0
-pyarrow<26.0.0,>=21.0.0
-click<9.0.0,>=8.2.0
-pydantic<3.0.0,>=2.12.0
-aiofiles<26.0.0,>=24.1.0
-rich<16.0.0,>=14.0.0
-cryptography<50.0.0,>=45.0.0
-PyYAML<7.0.0,>=6.0.2
-certbot<6.0.0,>=5.0.0
-numpy<3.0.0,>=2.2.0
-Pillow<13.0.0,>=11.3.0
-trainingsample<0.4.0,>=0.3.2
-pandas<4.0.0,>=3.0.0
-datasets<6.0.0,>=5.0.0
-boto3<2.0.0,>=1.43.0
-webshart<0.7.0,>=0.6.0
-pylance<9.0.0,>=8.0.0
-duckdb<2.0.0,>=1.5.0
-aiohttp<4.0.0,>=3.13.3
-fastapi<0.137.0,>=0.133.0
-uvicorn<1.0.0,>=0.35.0
-huggingface-hub<2.0.0,>=1.5.0
-psutil<8.0.0,>=7.0.0
-requests<3.0.0,>=2.32.0
-tqdm<5.0.0,>=4.67.0
-urwid<5.0.0,>=3.0.2
-
-[all]
-vllm<0.32.0,>=0.31.0
-torch==2.13.0
-torchvision==0.28.0
-torchaudio==2.11.0
-transformers<5.18.0,>=5.10.4
-qwen-vl-utils<0.1.0,>=0.0.14
-
-[apple]
-torch>=2.13.0
-torchvision>=0.28.0
-torchaudio>=2.11.0
-
-[apple:platform_system == "Darwin" and platform_machine == "arm64" and python_version >= "3.12"]
-vllm-metal==0.1.0
-
-[captioning]
-vllm<0.32.0,>=0.31.0
-torch==2.13.0
-torchvision==0.28.0
-torchaudio==2.11.0
-transformers<5.18.0,>=5.10.4
-qwen-vl-utils<0.1.0,>=0.0.14
-
-[cpu]
-torch>=2.13.0
-torchvision>=0.28.0
-torchaudio>=2.11.0
-
-[cuda]
-torch>=2.13.0
-torchvision>=0.28.0
-torchaudio>=2.11.0
-
-[cuda13]
-torch>=2.13.0
-torchvision>=0.28.0
-torchaudio>=2.11.0
-
-[dev]
-pytest<9.0.0,>=8.0.0
-pytest-asyncio<2.0.0,>=1.1.0
-pytest-cov<7.0.0,>=6.0.0
-transformers<5.18.0,>=5.10.4
-black<26.0.0,>=25.0.0
-ruff<1.0.0,>=0.12.0
-mypy<2.0.0,>=1.17.0
-
-[openai]
-
-[rocm]
-torch>=2.13.0
-torchvision>=0.28.0
-torchaudio>=2.11.0
-
-[vllm]
-vllm<0.32.0,>=0.31.0
-torch==2.13.0
-torchvision==0.28.0
-torchaudio==2.11.0
-transformers<5.18.0,>=5.10.4
-qwen-vl-utils<0.1.0,>=0.0.14