hyperstudy-perception 0.1.1__tar.gz → 0.1.3__tar.gz

{hyperstudy_perception-0.1.1 → hyperstudy_perception-0.1.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: hyperstudy-perception
-Version: 0.1.1
+Version: 0.1.3
 Summary: Video perception pipeline: shots, speech, faces and captions from a video, as a versioned track bundle, run locally, on cloud GPUs or lab servers
 Project-URL: Homepage, https://github.com/hyperstudyio/hyperstudy-perception
 Project-URL: Changelog, https://github.com/hyperstudyio/hyperstudy-perception/releases

{hyperstudy_perception-0.1.1 → hyperstudy_perception-0.1.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "hyperstudy-perception"
-version = "0.1.1"
+version = "0.1.3"
 description = "Video perception pipeline: shots, speech, faces and captions from a video, as a versioned track bundle, run locally, on cloud GPUs or lab servers"
 readme = "README.md"
 license = "MIT"

hyperstudy_perception-0.1.3/src/perception/_pipeline/RELEASE.json

@@ -0,0 +1,255 @@
+{
+"commit": "ba5bb86926a9a7218d908ad5f7dc0371cf87a0c3",
+"files": [
+"cloud/runpod/guard.sh",
+"cloud/runpod/image/Dockerfile",
+"cloud/runpod/image/start.sh",
+"cloud/runpod/image/tag.py",
+"cloud/skypilot/gpu-faces.yaml",
+"cloud/skypilot/gpu-tune.yaml",
+"cloud/skypilot/guard.sh",
+"cloud/skypilot/iam-perception-node.json",
+"cloud/skypilot/launch.sh",
+"cloud/skypilot/node-gpu.yaml",
+"cloud/skypilot/node-setup.sh",
+"cloud/skypilot/sky.yaml",
+"cloud/skypilot/smoke.yaml",
+"envs/api/pyproject.toml",
+"envs/api/uv.lock",
+"envs/asr-cuda/pyproject.toml",
+"envs/asr-cuda/uv.lock",
+"envs/audio-events/pyproject.toml",
+"envs/audio-events/uv.lock",
+"envs/audio/pyproject.toml",
+"envs/audio/uv.lock",
+"envs/core/pyproject.toml",
+"envs/core/uv.lock",
+"envs/faces/pyproject.toml",
+"envs/faces/uv.lock",
+"envs/gaze/pyproject.toml",
+"envs/gaze/uv.lock",
+"envs/prosody/pyproject.toml",
+"envs/prosody/uv.lock",
+"envs/vision/pyproject.toml",
+"envs/vision/uv.lock",
+"graphs/decode.yaml",
+"graphs/faces.yaml",
+"graphs/full.yaml",
+"graphs/mvp-captions.yaml",
+"graphs/mvp.yaml",
+"nodes/audio/asr/manifest.yaml",
+"nodes/audio/asr/nemo_model.py",
+"nodes/audio/asr/nemo_tokens.py",
+"nodes/audio/asr/node.py",
+"nodes/audio/asr/node_nemo.py",
+"nodes/audio/asr/prepare.py",
+"nodes/audio/asr/raw_tokens.py",
+"nodes/audio/asr/test_nemo_model.py",
+"nodes/audio/asr/test_nemo_tokens.py",
+"nodes/audio/asr/test_raw_tokens.py",
+"nodes/audio/asr/test_wav.py",
+"nodes/audio/asr/utterances.py",
+"nodes/audio/asr/wav.py",
+"nodes/audio/diarize/diarization_raw.py",
+"nodes/audio/diarize/diarizer.py",
+"nodes/audio/diarize/manifest.yaml",
+"nodes/audio/diarize/node.py",
+"nodes/audio/diarize/pcm16.py",
+"nodes/audio/diarize/prepare.py",
+"nodes/audio/diarize/test_diarization_raw.py",
+"nodes/audio/diarize/test_diarizer.py",
+"nodes/audio/diarize/test_pcm16.py",
+"nodes/audio/diarize/test_voices.py",
+"nodes/audio/diarize/voices.py",
+"nodes/audio/prosody-captions/captions_raw.py",
+"nodes/audio/prosody-captions/manifest.yaml",
+"nodes/audio/prosody-captions/node.py",
+"nodes/audio/prosody-captions/style_captions.py",
+"nodes/audio/prosody-captions/test_style_captions.py",
+"nodes/audio/prosody/manifest.yaml",
+"nodes/audio/prosody/node.py",
+"nodes/audio/prosody/odyssey.py",
+"nodes/audio/prosody/praat.py",
+"nodes/audio/prosody/prepare.py",
+"nodes/audio/prosody/prosody_layer.py",
+"nodes/audio/prosody/prosody_measure.py",
+"nodes/audio/prosody/prosody_tracks.py",
+"nodes/audio/prosody/prosody_units.py",
+"nodes/audio/prosody/test_prosody_layer.py",
+"nodes/audio/prosody/test_prosody_measure.py",
+"nodes/audio/prosody/test_prosody_tracks.py",
+"nodes/audio/prosody/test_prosody_units.py",
+"nodes/audio/prosody/third_party/README.md",
+"nodes/audio/prosody/third_party/odyssey/LICENSE",
+"nodes/audio/prosody/third_party/odyssey/ser_model.py",
+"nodes/audio/sounds/audioset_labels.csv",
+"nodes/audio/sounds/efficientat/LICENSE",
+"nodes/audio/sounds/efficientat/UPSTREAM.md",
+"nodes/audio/sounds/efficientat/__init__.py",
+"nodes/audio/sounds/efficientat/mn/__init__.py",
+"nodes/audio/sounds/efficientat/mn/attention_pooling.py",
+"nodes/audio/sounds/efficientat/mn/block_types.py",
+"nodes/audio/sounds/efficientat/mn/model.py",
+"nodes/audio/sounds/efficientat/mn/utils.py",
+"nodes/audio/sounds/efficientat/preprocess.py",
+"nodes/audio/sounds/manifest.yaml",
+"nodes/audio/sounds/models.py",
+"nodes/audio/sounds/node.py",
+"nodes/audio/sounds/panns/LICENSE.MIT",
+"nodes/audio/sounds/panns/UPSTREAM.md",
+"nodes/audio/sounds/panns/__init__.py",
+"nodes/audio/sounds/panns/cnn14.py",
+"nodes/audio/sounds/panns/pytorch_utils.py",
+"nodes/audio/sounds/prepare.py",
+"nodes/audio/sounds/sound_events.py",
+"nodes/audio/sounds/sound_scores.py",
+"nodes/audio/sounds/soundtrack.py",
+"nodes/audio/sounds/test_sound_events.py",
+"nodes/audio/sounds/test_sound_scores.py",
+"nodes/audio/sounds/test_soundtrack.py",
+"nodes/audio/waveform/manifest.yaml",
+"nodes/audio/waveform/node.py",
+"nodes/audio/waveform/test_waveform_peaks.py",
+"nodes/audio/waveform/waveform_peaks.py",
+"nodes/conftest.py",
+"nodes/core/decode/manifest.yaml",
+"nodes/core/decode/media.py",
+"nodes/core/decode/node.py",
+"nodes/core/decode/shot_scores.py",
+"nodes/core/decode/test_media.py",
+"nodes/core/decode/test_shot_scores.py",
+"nodes/testdata/README.md",
+"nodes/testdata/rotated-90.mp4",
+"nodes/vision/active-speaker/asd_crops.py",
+"nodes/vision/active-speaker/asd_raw.py",
+"nodes/vision/active-speaker/manifest.yaml",
+"nodes/vision/active-speaker/models.py",
+"nodes/vision/active-speaker/node.py",
+"nodes/vision/active-speaker/prepare.py",
+"nodes/vision/active-speaker/speaker_links.py",
+"nodes/vision/active-speaker/test_asd_crops.py",
+"nodes/vision/active-speaker/test_asd_raw.py",
+"nodes/vision/active-speaker/test_speaker_links.py",
+"nodes/vision/active-speaker/third_party/README.md",
+"nodes/vision/active-speaker/third_party/lr_asd/LICENSE",
+"nodes/vision/active-speaker/third_party/lr_asd/loss.py",
+"nodes/vision/active-speaker/third_party/lr_asd/model/Classifier.py",
+"nodes/vision/active-speaker/third_party/lr_asd/model/Encoder.py",
+"nodes/vision/active-speaker/third_party/lr_asd/model/Model.py",
+"nodes/vision/active-speaker/video25.py",
+"nodes/vision/body-pose/bodies.py",
+"nodes/vision/body-pose/detections.py",
+"nodes/vision/body-pose/keypoints.json",
+"nodes/vision/body-pose/manifest.yaml",
+"nodes/vision/body-pose/models.py",
+"nodes/vision/body-pose/node.py",
+"nodes/vision/body-pose/prepare.py",
+"nodes/vision/body-pose/test_bodies.py",
+"nodes/vision/body-pose/test_detections.py",
+"nodes/vision/face-track/face_detail.py",
+"nodes/vision/face-track/frames.py",
+"nodes/vision/face-track/loader.py",
+"nodes/vision/face-track/manifest.yaml",
+"nodes/vision/face-track/node.py",
+"nodes/vision/face-track/prepare.py",
+"nodes/vision/face-track/pyfeat_raw.py",
+"nodes/vision/face-track/test_face_detail.py",
+"nodes/vision/face-track/test_frames.py",
+"nodes/vision/face-track/test_loader.py",
+"nodes/vision/face-track/test_pyfeat_raw.py",
+"nodes/vision/face-track/test_thumbnails.py",
+"nodes/vision/face-track/thumbnails.py",
+"nodes/vision/face-track/tracking.py",
+"nodes/vision/facing/facing.py",
+"nodes/vision/facing/facing_raw.py",
+"nodes/vision/facing/manifest.yaml",
+"nodes/vision/facing/models.py",
+"nodes/vision/facing/node.py",
+"nodes/vision/facing/prepare.py",
+"nodes/vision/facing/test_facing.py",
+"nodes/vision/facing/test_facing_raw.py",
+"nodes/vision/gaze/gaze_polygons.py",
+"nodes/vision/gaze/gaze_targets.py",
+"nodes/vision/gaze/heatmaps.py",
+"nodes/vision/gaze/manifest.yaml",
+"nodes/vision/gaze/models.py",
+"nodes/vision/gaze/node.py",
+"nodes/vision/gaze/prepare.py",
+"nodes/vision/gaze/test_gaze_polygons.py",
+"nodes/vision/gaze/test_gaze_targets.py",
+"nodes/vision/gaze/test_heatmaps.py",
+"nodes/vision/gaze/third_party/README.md",
+"nodes/vision/gaze/third_party/dinov2/LICENSE",
+"nodes/vision/gaze/third_party/dinov2/dinov2/__init__.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/layers/__init__.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/layers/attention.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/layers/block.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/layers/dino_head.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/layers/drop_path.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/layers/layer_scale.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/layers/mlp.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/layers/patch_embed.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/layers/swiglu_ffn.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/models/__init__.py",
+"nodes/vision/gaze/third_party/dinov2/dinov2/models/vision_transformer.py",
+"nodes/vision/gaze/third_party/gazelle/LICENSE",
+"nodes/vision/gaze/third_party/gazelle/gazelle/backbone.py",
+"nodes/vision/gaze/third_party/gazelle/gazelle/model.py",
+"nodes/vision/gaze/third_party/gazelle/gazelle/utils.py",
+"nodes/vision/gestures/gesture_evidence.py",
+"nodes/vision/gestures/gesture_inputs.py",
+"nodes/vision/gestures/gesture_rules.py",
+"nodes/vision/gestures/manifest.yaml",
+"nodes/vision/gestures/node.py",
+"nodes/vision/gestures/test_gesture_evidence.py",
+"nodes/vision/gestures/test_gesture_rules.py",
+"nodes/vision/hands/hand_detections.py",
+"nodes/vision/hands/hand_tracks.py",
+"nodes/vision/hands/manifest.yaml",
+"nodes/vision/hands/models.py",
+"nodes/vision/hands/node.py",
+"nodes/vision/hands/prepare.py",
+"nodes/vision/hands/scipy_lsap/LICENSE",
+"nodes/vision/hands/scipy_lsap/UPSTREAM.md",
+"nodes/vision/hands/scipy_lsap/__init__.py",
+"nodes/vision/hands/scipy_lsap/rectangular_lsap.py",
+"nodes/vision/hands/test_hand_detections.py",
+"nodes/vision/hands/test_hand_tracks.py",
+"nodes/vision/hands/test_node.py",
+"nodes/vision/hands/test_rectangular_lsap.py",
+"nodes/vision/objects-link/manifest.yaml",
+"nodes/vision/objects-names/manifest.yaml",
+"nodes/vision/objects-track/manifest.yaml",
+"nodes/vision/objects/budget.py",
+"nodes/vision/objects/garments.py",
+"nodes/vision/objects/geometry.py",
+"nodes/vision/objects/haiku.py",
+"nodes/vision/objects/link.py",
+"nodes/vision/objects/linking.py",
+"nodes/vision/objects/models.py",
+"nodes/vision/objects/names.py",
+"nodes/vision/objects/object_raw.py",
+"nodes/vision/objects/object_tracks.py",
+"nodes/vision/objects/person_masks.py",
+"nodes/vision/objects/prepare.py",
+"nodes/vision/objects/shot_frames.py",
+"nodes/vision/objects/test_garments.py",
+"nodes/vision/objects/test_models_hub.py",
+"nodes/vision/objects/test_object_budget.py",
+"nodes/vision/objects/test_object_links.py",
+"nodes/vision/objects/test_object_raw.py",
+"nodes/vision/objects/test_person_masks.py",
+"nodes/vision/objects/test_shot_frames.py",
+"nodes/vision/objects/test_split.py",
+"nodes/vision/objects/testdata/node_1_1_0.py",
+"nodes/vision/objects/track.py",
+"nodes/vision/objects/vocabulary.py",
+"nodes/vlm/shot-captions/manifest.yaml",
+"nodes/vlm/shot-captions/node.py",
+"schema/README.md",
+"schema/bundle-0.1.schema.json",
+"sdk/pyproject.toml",
+"sdk/src/perception_node/__init__.py",
+"sdk/src/perception_node/weights.py"
+]
+}

{hyperstudy_perception-0.1.1 → hyperstudy_perception-0.1.3}/src/perception/_pipeline/cloud/skypilot/gpu-tune.yaml

@@ -43,7 +43,7 @@ run: |
   uv run perception run ../graphs/decode.yaml --input video=/data/clip.mp4
   for point in $SWEEP; do
     b=${point%:*}; w=${point#*:}
-    sed "s/uses: vision.face-track@1.0.0/&\n    config: {batch_size: $b, num_workers: $w}/" ../graphs/faces.yaml > ../graphs/tune-$b-$w.yaml
+    sed "s/uses: vision.face-track@1.4.0/&\n    config: {batch_size: $b, num_workers: $w}/" ../graphs/faces.yaml > ../graphs/tune-$b-$w.yaml
     echo "=== batch $b workers $w $(date +%T)" | tee -a ~/out/sweep.txt
     SECONDS=0
     # A failed point is recorded (with its error code, from status) and the sweep goes on.

hyperstudy_perception-0.1.3/src/perception/_pipeline/envs/asr-cuda/pyproject.toml

@@ -0,0 +1,29 @@
+[project]
+name = "perception-env-asr-cuda"
+version = "0.1.0"
+description = "Speech transcript on an NVIDIA GPU or a CPU: Parakeet TDT 0.6B v3 through NVIDIA NeMo (audio.asr's node_nemo.py)"
+requires-python = ">=3.12,<3.13"
+# NeMo 3.0.0, the current release, with its ASR extra; torch 2.12.0, the version NeMo 3.0.0 pins for its own CUDA
+# builds (its cu12 and cu13 extras). PyPI's torch 2.12.0 for Linux is built for CUDA 13.0, whose runtime, cuBLAS and
+# cuDNN 9.20 come with it as wheels: an NVIDIA driver >= 580 (as RunPod's pods have, cloud.py). NeMo's own extras pull
+# torch from PyTorch's indexes and add numba-cuda and Transformer Engine for training; inference needs neither.
+# A lock of its own: NeMo's ASR extra caps lightning at 2.4.0 (envs/audio's pyannote 4.0.7 runs on lightning 2.6.6,
+# and torch 2.8.0 for CUDA 12.8) and brings hydra, omegaconf, lhotse, librosa 1.0 and transformers 5.21 (envs/gaze has
+# 5.20.2 under SAM 3): either shared lock would move versions under the nodes measured on them.
+# pyarrow writes the tokens side file (raw_tokens.py), pinned exactly as in envs/audio: its version names itself in the
+# file's footer.
+dependencies = [
+  "perception-node",
+  "nemo-toolkit[asr]==3.0.0",
+  "torch==2.12.0",
+  "pyarrow==26.0.0",
+]
+
+[tool.uv]
+package = false
+# The runners are linux/amd64 (CUDA, and CPU). darwin/arm64 resolves too, so the node's code can be run on a Mac's CPU
+# from this same lock (it was, on OOO): the manifest never places it there, where the MLX runner is the node's
+environments = ["sys_platform == 'linux' and platform_machine == 'x86_64'", "sys_platform == 'darwin' and platform_machine == 'arm64'"]
+
+[tool.uv.sources]
+perception-node = { path = "../../sdk", editable = true }