I have a gpu! I want to use it!

.vscode/settings.json

@@ -5,5 +5,5 @@
     "**/*.rpymc": true,
     "**/cache/": true
   },
-  "pylint.args": ["--generated-members", "cv2.*"]
+  "pylint.args": ["--generated-members", "cv2.*"],
 }

main.py

@@ -74,8 +74,8 @@ def main():
         cigarette_model=cigarette_model,
         verifier_model=verifier,
         vlm_model=None,  # Optional
-        store=alert_store,
+        sinks=[video_sink, alert_store],
-        sinks=[video_sink],
+        inference_interval=0,
     )
 
     if args.rtsp:
@@ -92,7 +92,7 @@ def main():
         print("Stopped by user.")
     finally:
         if args.rtsp and hasattr(source, "stop"):
-            source.stop()
+            source.stop() # pyright: ignore[reportAttributeAccessIssue]
         if video_sink:
             video_sink.close()
 

pipeline/core.py

@@ -1,20 +1,13 @@
 import logging
 from collections import deque
-from typing import Optional
+from typing import List, Optional, Tuple
 
-from pipeline.protocols import AlertStore, Model, StreamSource, Verifier
+from pipeline.protocols import Model, Sink, StreamSource, Verifier
-from pipeline.types import Frame, Ticket
+from pipeline.types import BoundingBox, Detection, Frame, Ticket
 
 logger = logging.getLogger(__name__)
 
 
-class ConsoleStore:
-    def store(self, ticket: Ticket):
-        logger.info(
-            f"[STORE] Persisting Alert! Conf: {ticket.confidence:.2f} @ Stage: {ticket.stage_confirmed}"
-        )
-
-
 class SmokingDetectionPipeline:
     def __init__(
         self,
@@ -22,19 +15,20 @@ class SmokingDetectionPipeline:
         cigarette_model: Model,
         verifier_model: Verifier,
         vlm_model: Optional[Verifier] = None,
-        store: Optional[AlertStore] = None,
+        sinks: Optional[list[Sink]] = None,
-        sinks: Optional[list] = None,
+        inference_interval: float = 0.5,
     ):
         self.person_model = person_model
         self.cigarette_model = cigarette_model
         self.verifier_model = verifier_model
         self.vlm_model = vlm_model
-        self.store = store or ConsoleStore()
         self.sinks = sinks or []
+        self.inference_interval = inference_interval
 
+        self.last_inference_time = 0.0
         # Buffer for temporal context (optional, for VLM or sequence logic)
         self.frame_buffer = deque(maxlen=30)
-        logger.info("Pipeline initialized.")
+        logger.info(f"Pipeline initialized with {len(self.sinks)} sinks.")
 
     def process_stream(self, source: StreamSource):
         """Main loop processing the video stream."""
@@ -47,7 +41,11 @@ class SmokingDetectionPipeline:
             if count % 100 == 0:
                 logger.info(f"Processed {count} frames...")
             self.frame_buffer.append(frame)
-            self._process_frame(frame)
+
+            # Skip inference if too soon
+            if (frame.timestamp - self.last_inference_time) >= self.inference_interval:
+                self._process_frame(frame)
+                self.last_inference_time = frame.timestamp
 
             # Update sinks (Visualizers, etc)
             for sink in self.sinks:
@@ -57,11 +55,7 @@ class SmokingDetectionPipeline:
 
     def _process_frame(self, frame: Frame):
         # --- Step 1: Broad Phase (Person Detection) ---
-        # "Is there a human here?"
-        # optimization: Resize or skip frames if inference is slow
         people = self.person_model.predict(frame.data)
-
-        # Add to frame detections for visualization
         frame.detections.extend(people)
 
         if not people:
@@ -70,29 +64,49 @@ class SmokingDetectionPipeline:
         logger.debug(f"Frame {frame.frame_id}: Found {len(people)} people.")
 
         # --- Step 2: Narrow Phase (Cigarette Candidate) ---
-        # "Is the human holding something small?"
+        candidates = self._find_candidates(frame, people)
-        # We could optimize by only running on crops of the 'people'
-        candidates = []
-        for person in people:
-            # ROI: Region of Interest (e.g., upper body or hands)
-            # EXPANSTION: Expand the person box to include extended arms/surroundings
-            h_img, w_img = frame.data.shape[:2]
-            margin = 0.2  # 20% expansion
 
+        if not candidates:
+            return  # EXIT
+
+        logger.debug(f"Frame {frame.frame_id}: Found {len(candidates)} candidates.")
+
+        # --- Step 3: Verification (Classifier) ---
+        best_hit = self._verify_candidates(frame, candidates)
+
+        # --- Step 4: Action ---
+        if best_hit:
+            person, cig, score = best_hit
+            ticket = Ticket(
+                frame_id=frame.frame_id,
+                confidence=score,
+                evidence=[frame.data],
+                stage_confirmed="Verifier",  # Simplified for now
+            )
+            frame.ticket = ticket
+
+    def _find_candidates(
+        self, frame: Frame, people: List[Detection]
+    ) -> List[Tuple[Detection, List[Detection], Tuple[int, int]]]:
+        """Runs cigarette detection on expanded person crops."""
+        candidates = []
+        h_img, w_img = frame.data.shape[:2]
+        margin = 0.2  # 20% expansion
+
+        for person in people:
+            # 1. Expand Box
             pb = person.box
             bw = pb.x2 - pb.x1
             bh = pb.y2 - pb.y1
 
-            # Expanded coordinates for Step 2
             ex1 = max(0, int(pb.x1 - bw * margin))
             ey1 = max(0, int(pb.y1 - bh * margin))
             ex2 = min(w_img, int(pb.x2 + bw * margin))
             ey2 = min(h_img, int(pb.y2 + bh * margin))
 
-            # Use expanded crop for cigarette detection
+            # 2. Crop
             person_crop = frame.data[ey1:ey2, ex1:ex2]
 
-            # If crop is too small, skip
             if (
                 person_crop.size == 0
                 or person_crop.shape[0] < 10
@@ -100,10 +114,10 @@ class SmokingDetectionPipeline:
             ):
                 continue
 
+            # 3. Detect
             cig_candidates = self.cigarette_model.predict(person_crop)
 
-            # Filter out extremely small candidates (noise)
+            # 4. Filter Noise
-            # User requested check similar to person_crop check (< 10px)
             cig_candidates = [
                 c
                 for c in cig_candidates
@@ -111,25 +125,10 @@ class SmokingDetectionPipeline:
             ]
 
             if cig_candidates:
-                # Store expansion offset for later coordinate translation
-                # We store (ex1, ey1) as metadata on the person detection or passed tuple?
-                # Easiest is to pass it in the tuple: (person, cigs, offset)
                 candidates.append((person, cig_candidates, (ex1, ey1)))
 
-                # Check detections (translate to global coordinates)
+                # 5. Global Coordinate Mapping (For Visualization)
                 for cig in cig_candidates:
-                    # Clone or modify the cig detection to be global
-                    # Assuming cig.box is relevant to person_crop
-                    # Global X = person_crop_origin_x + local_x
-                    # Global Y = person_crop_origin_y + local_y
-
-                    # Update in place might affect other logic if it assumes local,
-                    # but logic below (Expansion) seems to rely on cig.box being used on raw_person_crop
-                    # which implies local coordinates.
-                    # So we should create a COPY for visualization.
-
-                    from pipeline.types import BoundingBox, Detection
-
                     global_cig = Detection(
                         box=BoundingBox(
                             x1=ex1 + cig.box.x1,
@@ -144,97 +143,47 @@ class SmokingDetectionPipeline:
                     )
                     frame.detections.append(global_cig)
 
-        if not candidates:
+        return candidates
-            return  # EXIT
 
-        logger.debug(f"Frame {frame.frame_id}: Found {len(candidates)} candidates.")
+    def _verify_candidates(
-
+        self, frame: Frame, candidates: List[Tuple]
-        # --- Step 3: Verification (Classifier) ---
+    ) -> Optional[Tuple]:
-        # "Is that small thing actually a cigarette?"
+        """Runs verification on candidate crops."""
         confirmed_hits = []
+
         for person, cigs, (ex1, ey1) in candidates:
             for cig in cigs:
-                # Crop just the object.
+                # 1. Derive Crop Coordinates
-                # Note: box coordinates are relative to the person crop, or global?
-                # Ideally, detections should be unified.
-                # If cigarette_model returns boxes relative to person_crop, we need to adjust them.
-                # For simplicity in this mock, let's assume models return GLOBAL coords or handle their own relative logic wrapping.
-                # However, usually crop inference returns local coords.
-                # Let's assume for now the cig model knows how to return global.
-                # OR we pass the crop to verifier.
-
-                # IMPORTANT: If cigarette model ran on a crop, its output box is relative to that crop.
-                # We need to map it back or just use the crop directly?
-                # Actually, we likely want to crop the cigarette from the person crop.
-
-                # Simplified: Just rely on the verifier taking the small patch.
-
-                # To be robust, let's assume the Detector returns a Detection with a convenient 'crop' method or we handle coordinate translation.
-                # But for this MVP, let's assume we pass the crop derived from the box.
-
-                # FIX: If cig_candidates came from person_crop, their boxes are relative to (0,0) of person_crop.
-                # We should probably run verification on the crop defined by cig.box within person_crop.
-
-                # Let's grab the actual image patch for verification
-                # We can'teasily re-crop from global frame without translation.
-                # Strategy: cig detection on person_crop is easiest.
-
-                # Re-cutting the crop:
-                # We need the SAME crop used for detection, or at least the same coordinate system.
-                # Since we used the expanded crop (person_crop created via ex1, ey1),
-                # cig.box is relative to that.
-
-                # We can just re-extract the crop from the frame using ex1, ey1 if we didn't save it.
-                # Or just use the global coordinates we know now.
-
-                # Let's derive the object crop from the expanded person crop logic.
-
-                # --- Expansion Logic ---
-                # Expand the box to give context
-                # Here we are expanding the CIGARETTE box within the PERSON crop?
-                # The user asked to expand the hitbox (of the human) stage two too.
-                # We did that in Step 2.
-
-                # Now we need to prepare the crop for VERIFICATION (Step 3).
-                # We take the cig box (relative to expanded person crop) and add context?
-
                 cig_w = cig.box.x2 - cig.box.x1
                 cig_h = cig.box.y2 - cig.box.y1
 
-                pad_x = int(cig_w * 0.5)  # Add 50% width padding
+                pad_x = int(cig_w * 0.5)
-                pad_y = int(cig_h * 0.5)  # Add 50% height padding
+                pad_y = int(cig_h * 0.5)
 
-                # Coordinates relative to the expanded person crop
+                global_cx1 = ex1 + max(0, cig.box.x1 - pad_x)
-                cx1 = max(0, cig.box.x1 - pad_x)
+                global_cy1 = ey1 + max(0, cig.box.y1 - pad_y)
-                cy1 = max(0, cig.box.y1 - pad_y)
-                # We don't know the exact W/H of the expanded crop easily here without recomputing or checking boundaries
-                # But we can map to Global and crop from frame directly to be safe.
-
-                global_cx1 = ex1 + cx1
-                global_cy1 = ey1 + cy1
                 global_cx2 = ex1 + cig.box.x2 + pad_x
                 global_cy2 = ey1 + cig.box.y2 + pad_y
 
+                # Check bounds
                 h_img, w_img = frame.data.shape[:2]
-
                 final_x1 = max(0, global_cx1)
                 final_y1 = max(0, global_cy1)
                 final_x2 = min(w_img, global_cx2)
                 final_y2 = min(h_img, global_cy2)
 
+                # 2. Crop
                 obj_crop = frame.data[final_y1:final_y2, final_x1:final_x2]
 
                 if obj_crop.size == 0:
                     continue
 
+                # 3. Verify
                 score = self.verifier_model.verify(obj_crop)
 
-                if score > 0.8:  # High confidence threshold
+                if score > 0.8:  # High confidence
                     confirmed_hits.append((person, cig, score))
 
-                    # Add verified detection to frame for visualization
-                    from pipeline.types import BoundingBox, Detection
-
                     verified_det = Detection(
                         box=BoundingBox(
                             x1=ex1 + cig.box.x1,
@@ -249,23 +198,6 @@ class SmokingDetectionPipeline:
                     )
                     frame.detections.append(verified_det)
 
-                elif score > 0.4 and self.vlm_model:
-                    # Ambiguous case: Send to VLM?
-                    vlm_score = self.vlm_model.verify(
-                        obj_crop
-                    )  # Or pass seq of frames?
-                    if vlm_score > 0.8:
-                        confirmed_hits.append((person, cig, vlm_score))
-        # --- Step 4: Action ---
         if confirmed_hits:
-            best_hit = max(confirmed_hits, key=lambda x: x[2])
+            return max(confirmed_hits, key=lambda x: x[2])
-            ticket = Ticket(
+        return None
-                frame_id=frame.frame_id,
-                confidence=best_hit[2],
-                evidence=[frame.data],  # Store frame reference
-                stage_confirmed="VLM"
-                if self.vlm_model
-                and best_hit[2] == 0.1  # Logic here is a bit shaky from mock
-                else "Verifier",
-            )
-            self.store.store(ticket)

pipeline/models.py

@@ -1,6 +1,7 @@
 import logging
 from typing import Any, List
 
+import torch
 from ultralytics.models import YOLO
 
 from pipeline.protocols import Model, Verifier
@@ -8,6 +9,13 @@ from pipeline.types import BoundingBox, Detection
 
 logger = logging.getLogger(__name__)
 
+# Auto-detect device
+DEVICE = 0 if torch.cuda.is_available() else "cpu"
+if DEVICE == 0:
+    logger.info(f"GPU detected! Using CUDA device: {torch.cuda.get_device_name(0)}")
+else:
+    logger.info("No GPU detected. Running on CPU.")
+
 
 class YoloPersonDetector(Model):
     def __init__(self, model_path: str = "yolo11n.pt"):  # Using small model for speed
@@ -16,7 +24,9 @@ class YoloPersonDetector(Model):
 
     def predict(self, input_data: Any) -> List[Detection]:
         # input_data: numpy array (H, W, C)
-        results = self.model(input_data, classes=[0], verbose=False)  # 0 is person
+        results = self.model(
+            input_data, classes=[0], verbose=False, device=DEVICE
+        )  # 0 is person
 
         detections = []
         for r in results:
@@ -42,7 +52,7 @@ class YoloCigaretteDetector(Model):
     def predict(self, input_data: Any) -> List[Detection]:
         # This model should be trained on cigarette class
         # Assuming class 0 is cigarette for this custom model
-        results = self.model(input_data, verbose=False)
+        results = self.model(input_data, verbose=False, device=DEVICE)
 
         detections = []
         for r in results:
@@ -72,7 +82,7 @@ class YoloClassifier(Verifier):
         if input_data.size == 0:
             return 0.0
 
-        results = self.model(input_data, verbose=False)
+        results = self.model(input_data, verbose=False, device=DEVICE)
 
         # Check if we got any results
         if not results:

pipeline/outputs.py

@@ -6,7 +6,7 @@ from typing import List, Optional
 import cv2
 import numpy as np
 
-from pipeline.types import Detection, Frame, Ticket
+from pipeline.types import Detection, Frame
 
 logger = logging.getLogger(__name__)
 
@@ -19,10 +19,14 @@ class FileAlertStore:
 
     def __init__(self, path: str):
         self.path = path
-        self.file = open(path, "a")  # Append mode
+        self.file = open(path, "a", encoding="utf-8")  # Append mode
         logger.info(f"Initialized FileAlertStore writing to {path}")
 
-    def store(self, ticket: Ticket):
+    def handle(self, frame: Frame):
+        if not frame.ticket:
+            return
+
+        ticket = frame.ticket
         record = {
             "type": "alert",
             "frame_id": ticket.frame_id,

pipeline/protocols.py

@@ -1,6 +1,6 @@
 from typing import Any, Iterator, List, Protocol
 
-from pipeline.types import Detection, Frame, Ticket
+from pipeline.types import Detection, Frame
 
 
 class Model(Protocol):
@@ -21,7 +21,7 @@ class StreamSource(Protocol):
         ...
 
 
-class AlertStore(Protocol):
+class Sink(Protocol):
-    def store(self, ticket: Ticket):
+    def handle(self, frame: Frame):
-        """Persist the ticket."""
+        """Process or persist a frame (and its ticket/detections)."""
         ...

pipeline/types.py

@@ -1,5 +1,5 @@
 from dataclasses import dataclass, field
-from typing import Any, List
+from typing import Any, List, Optional
 
 import numpy as np
 
@@ -39,16 +39,6 @@ class Detection:
     metadata: dict = field(default_factory=dict)
 
 
-@dataclass
-class Frame:
-    """Represents a single frame in the video stream."""
-
-    timestamp: float
-    data: np.ndarray  # numpy array (OpenCV image)
-    frame_id: int
-    detections: List[Detection] = field(default_factory=list)
-
-
 @dataclass
 class Ticket:
     """Represents a potential alert event."""
@@ -58,3 +48,14 @@ class Ticket:
     evidence: List[Any]  # List of images or crops
     stage_confirmed: str
     metadata: dict = field(default_factory=dict)
+
+
+@dataclass
+class Frame:
+    """Represents a single frame in the video stream."""
+
+    timestamp: float
+    data: np.ndarray  # numpy array (OpenCV image)
+    frame_id: int
+    detections: List[Detection] = field(default_factory=list)
+    ticket: Optional[Ticket] = None

pyproject.toml

@@ -7,5 +7,19 @@ requires-python = ">=3.14"
 dependencies = [
     "numpy>=2.4.0",
     "opencv-python>=4.11.0.86",
+    "torch>=2.9.1",
     "ultralytics>=8.3.248",
 ]
+
+[[tool.uv.index]]
+name = "pytorch-cu130"
+url = "https://download.pytorch.org/whl/cu130"
+explicit = true
+
+[tool.uv.sources]
+torch = [
+  { index = "pytorch-cu130", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+]
+torchvision = [
+  { index = "pytorch-cu130", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+]

uv.lock

@@ -5,7 +5,8 @@ resolution-markers = [
     "sys_platform == 'win32'",
     "sys_platform == 'darwin'",
     "platform_machine == 'aarch64' and sys_platform == 'linux'",
-    "(platform_machine != 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "platform_machine != 'aarch64' and sys_platform == 'linux'",
+    "sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
 ]
 
 [[package]]
@@ -134,6 +135,8 @@ source = { virtual = "." }
 dependencies = [
     { name = "numpy" },
     { name = "opencv-python" },
+    { name = "torch", version = "2.9.1", source = { registry = "https://pypi.org/simple" }, marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
+    { name = "torch", version = "2.9.1+cu130", source = { registry = "https://download.pytorch.org/whl/cu130" }, marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
     { name = "ultralytics" },
 ]
 
@@ -141,6 +144,8 @@ dependencies = [
 requires-dist = [
     { name = "numpy", specifier = ">=2.4.0" },
     { name = "opencv-python", specifier = ">=4.11.0.86" },
+    { name = "torch", marker = "sys_platform != 'linux' and sys_platform != 'win32'", specifier = ">=2.9.1" },
+    { name = "torch", marker = "sys_platform == 'linux' or sys_platform == 'win32'", specifier = ">=2.9.1", index = "https://download.pytorch.org/whl/cu130" },
     { name = "ultralytics", specifier = ">=8.3.248" },
 ]
 
@@ -310,137 +315,152 @@ wheels = [
 ]
 
 [[package]]
-name = "nvidia-cublas-cu12"
+name = "nvidia-cublas"
-version = "12.8.4.1"
+version = "13.0.0.19"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/dc/61/e24b560ab2e2eaeb3c839129175fb330dfcfc29e5203196e5541a4c44682/nvidia_cublas_cu12-12.8.4.1-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:8ac4e771d5a348c551b2a426eda6193c19aa630236b418086020df5ba9667142", size = 594346921, upload-time = "2025-03-07T01:44:31.254Z" },
+    { url = "https://files.pythonhosted.org/packages/02/99/8447b9ee9f070522ee66604ee819d632ab4568c68b3134cebd3837a015cd/nvidia_cublas-13.0.0.19-py3-none-manylinux_2_27_aarch64.whl", hash = "sha256:381b1a0ca636fdcb6920a871e8fc89dbfd1f6157f421ed0a6f2673e14cffd3bd", size = 539001158, upload-time = "2025-08-04T10:19:50.761Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/99/210e113dde53955e97042bd76dc4ad927eca04c5b4645ec157cc59f4f3ae/nvidia_cublas-13.0.0.19-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:f6723af2e8e2600a11dc384037d90d9bf93070e346c24ef2e8f9001658c99896", size = 419392356, upload-time = "2025-08-04T10:20:19.449Z" },
 ]
 
 [[package]]
-name = "nvidia-cuda-cupti-cu12"
+name = "nvidia-cuda-cupti"
-version = "12.8.90"
+version = "13.0.48"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/f8/02/2adcaa145158bf1a8295d83591d22e4103dbfd821bcaf6f3f53151ca4ffa/nvidia_cuda_cupti_cu12-12.8.90-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:ea0cb07ebda26bb9b29ba82cda34849e73c166c18162d3913575b0c9db9a6182", size = 10248621, upload-time = "2025-03-07T01:40:21.213Z" },
+    { url = "https://files.pythonhosted.org/packages/72/63/e9c12c3ae07c1f3a0821536bc188d7bf76e1b633b3bcd2bd393b00bb3426/nvidia_cuda_cupti-13.0.48-py3-none-manylinux_2_25_aarch64.whl", hash = "sha256:67c22627ef436afcf080b48e4ad17b3f83d9e7c0d990ad0c6c0627b01fb92ccc", size = 10171189, upload-time = "2025-08-04T10:16:24.39Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/28/e37d62ff27b4462953fdd5713d8a78760578dfa12685c30b71b55fab57b1/nvidia_cuda_cupti-13.0.48-py3-none-manylinux_2_25_x86_64.whl", hash = "sha256:417699e216b23d81bc0bbcb7032352f81b9c5372ef73c097a01abb83125a3d09", size = 10718148, upload-time = "2025-08-04T10:16:33.605Z" },
 ]
 
 [[package]]
-name = "nvidia-cuda-nvrtc-cu12"
+name = "nvidia-cuda-nvrtc"
-version = "12.8.93"
+version = "13.0.48"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/05/6b/32f747947df2da6994e999492ab306a903659555dddc0fbdeb9d71f75e52/nvidia_cuda_nvrtc_cu12-12.8.93-py3-none-manylinux2010_x86_64.manylinux_2_12_x86_64.whl", hash = "sha256:a7756528852ef889772a84c6cd89d41dfa74667e24cca16bb31f8f061e3e9994", size = 88040029, upload-time = "2025-03-07T01:42:13.562Z" },
+    { url = "https://files.pythonhosted.org/packages/be/5b/f7636b3d66caefade6a0a0dc5b705c259a2062c20ad18b432b3129d348e0/nvidia_cuda_nvrtc-13.0.48-py3-none-manylinux2010_x86_64.manylinux_2_12_x86_64.whl", hash = "sha256:87e13d186905a35e7c04ad553a2abded0fba22f93b43d02e5da6f6cf73fb4d0a", size = 90214268, upload-time = "2025-08-04T10:18:09.305Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/bd/eb18593b43dae42312612ffbac24b8e68149e590102c3b6cc2e3d3792069/nvidia_cuda_nvrtc-13.0.48-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:6ccf1ef1b90a0763ac7536f3c17046659d89869d76b98ac358efc2e09b348365", size = 43013627, upload-time = "2025-08-04T10:17:57.338Z" },
 ]
 
 [[package]]
-name = "nvidia-cuda-runtime-cu12"
+name = "nvidia-cuda-runtime"
-version = "12.8.90"
+version = "13.0.48"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/0d/9b/a997b638fcd068ad6e4d53b8551a7d30fe8b404d6f1804abf1df69838932/nvidia_cuda_runtime_cu12-12.8.90-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:adade8dcbd0edf427b7204d480d6066d33902cab2a4707dcfc48a2d0fd44ab90", size = 954765, upload-time = "2025-03-07T01:40:01.615Z" },
+    { url = "https://files.pythonhosted.org/packages/55/3b/c5e5d8aafd355e2ff9922472ba71251331af6cc866e5b04a3b1dc8f58977/nvidia_cuda_runtime-13.0.48-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:b807c0bb925a307bfa667a24f24d253aef8eda3ac4be66b333f2c9d357557008", size = 2260687, upload-time = "2025-08-04T10:15:41.292Z" },
+    { url = "https://files.pythonhosted.org/packages/cc/78/edb119083ca2ff0f09ab0cd597e97775ac3f575b8aa0caf10d68ed49e032/nvidia_cuda_runtime-13.0.48-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:5b54d12087a1abff81a4cbfa6556876e3afea1fc60da2e0816da374619810c89", size = 2242632, upload-time = "2025-08-04T10:15:49.339Z" },
 ]
 
 [[package]]
-name = "nvidia-cudnn-cu12"
+name = "nvidia-cudnn-cu13"
-version = "9.10.2.21"
+version = "9.13.0.50"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "nvidia-cublas-cu12", marker = "(platform_machine != 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32')" },
+    { name = "nvidia-cublas", marker = "sys_platform == 'linux'" },
 ]
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/ba/51/e123d997aa098c61d029f76663dedbfb9bc8dcf8c60cbd6adbe42f76d049/nvidia_cudnn_cu12-9.10.2.21-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:949452be657fa16687d0930933f032835951ef0892b37d2d53824d1a84dc97a8", size = 706758467, upload-time = "2025-06-06T21:54:08.597Z" },
+    { url = "https://files.pythonhosted.org/packages/8a/9c/9e99c00dc23db324244ec257d1e84d79539202ee2f185dee2c1fa97c9549/nvidia_cudnn_cu13-9.13.0.50-py3-none-manylinux_2_27_aarch64.whl", hash = "sha256:33f0aa0b64230101b348648fd0693342188071d3f8a137c0cf50051c24b3584b", size = 412337597, upload-time = "2025-09-04T20:22:31.535Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/68/2712854561170b2a81bea7b6b35cc1ae264d9794c0c218986e5c685d45f7/nvidia_cudnn_cu13-9.13.0.50-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:2150b4850725d30653ec3e365f0732e3e2e3eb8633cf3bd2d3117628dea8b4f9", size = 348571624, upload-time = "2025-09-04T20:23:26.544Z" },
 ]
 
 [[package]]
-name = "nvidia-cufft-cu12"
+name = "nvidia-cufft"
-version = "11.3.3.83"
+version = "12.0.0.15"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "nvidia-nvjitlink-cu12", marker = "(platform_machine != 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32')" },
+    { name = "nvidia-nvjitlink", marker = "sys_platform == 'linux'" },
 ]
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/1f/13/ee4e00f30e676b66ae65b4f08cb5bcbb8392c03f54f2d5413ea99a5d1c80/nvidia_cufft_cu12-11.3.3.83-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:4d2dd21ec0b88cf61b62e6b43564355e5222e4a3fb394cac0db101f2dd0d4f74", size = 193118695, upload-time = "2025-03-07T01:45:27.821Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/e9/4e49b1baf6899e42eeec324a49d7aa2219fec42076327c4e468000dd375a/nvidia_cufft-12.0.0.15-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:1885731254835797572ff075f3daf43a2a0a2801210dea26971940dae7e1a367", size = 214053580, upload-time = "2025-08-04T10:20:45.781Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/9f/e298b66e584ad25bd78ad4a45b061fe7bb57a1ec011128089404ce3fcc7d/nvidia_cufft-12.0.0.15-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:9f160b1f018e80bcb0d7c0fa50564b042fa26b13edc1b1ff14b6375a9edd2812", size = 214085489, upload-time = "2025-08-04T10:21:02.975Z" },
 ]
 
 [[package]]
-name = "nvidia-cufile-cu12"
+name = "nvidia-cufile"
-version = "1.13.1.3"
+version = "1.15.0.42"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/bb/fe/1bcba1dfbfb8d01be8d93f07bfc502c93fa23afa6fd5ab3fc7c1df71038a/nvidia_cufile_cu12-1.13.1.3-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:1d069003be650e131b21c932ec3d8969c1715379251f8d23a1860554b1cb24fc", size = 1197834, upload-time = "2025-03-07T01:45:50.723Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/0a/4adf0c9bb1241cd1314fc923fde00f3749c7fc785b1e3b3f4a104cd3090c/nvidia_cufile-1.15.0.42-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:c8f9813eff24d61586699c615e39817e2b4e4f642cace32733c2ab6f663a7eab", size = 1223104, upload-time = "2025-08-04T10:21:31.131Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/a5/636baa43399ea10d22b63e7454f22a92ace4a7eaa3c45b94607250857e2d/nvidia_cufile-1.15.0.42-py3-none-manylinux_2_27_aarch64.whl", hash = "sha256:bced4036b5a8dbf57e4d78cd4fafefec58ad754b784a9eaa272b011896754c62", size = 1136527, upload-time = "2025-08-04T10:21:22.441Z" },
 ]
 
 [[package]]
-name = "nvidia-curand-cu12"
+name = "nvidia-curand"
-version = "10.3.9.90"
+version = "10.4.0.35"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/fb/aa/6584b56dc84ebe9cf93226a5cde4d99080c8e90ab40f0c27bda7a0f29aa1/nvidia_curand_cu12-10.3.9.90-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:b32331d4f4df5d6eefa0554c565b626c7216f87a06a4f56fab27c3b68a830ec9", size = 63619976, upload-time = "2025-03-07T01:46:23.323Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/72/7c2ae24fb6b63a32e6ae5d241cc65263ea18d08802aaae087d9f013335a2/nvidia_curand-10.4.0.35-py3-none-manylinux_2_27_aarch64.whl", hash = "sha256:133df5a7509c3e292aaa2b477afd0194f06ce4ea24d714d616ff36439cee349a", size = 61962106, upload-time = "2025-08-04T10:21:41.128Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/9f/be0a41ca4a4917abf5cb9ae0daff1a6060cc5de950aec0396de9f3b52bc5/nvidia_curand-10.4.0.35-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:1aee33a5da6e1db083fe2b90082def8915f30f3248d5896bcec36a579d941bfc", size = 59544258, upload-time = "2025-08-04T10:22:03.992Z" },
 ]
 
 [[package]]
-name = "nvidia-cusolver-cu12"
+name = "nvidia-cusolver"
-version = "11.7.3.90"
+version = "12.0.3.29"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "nvidia-cublas-cu12", marker = "(platform_machine != 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32')" },
+    { name = "nvidia-cublas", marker = "sys_platform == 'linux'" },
-    { name = "nvidia-cusparse-cu12", marker = "(platform_machine != 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32')" },
+    { name = "nvidia-cusparse", marker = "sys_platform == 'linux'" },
-    { name = "nvidia-nvjitlink-cu12", marker = "(platform_machine != 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32')" },
+    { name = "nvidia-nvjitlink", marker = "sys_platform == 'linux'" },
 ]
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/85/48/9a13d2975803e8cf2777d5ed57b87a0b6ca2cc795f9a4f59796a910bfb80/nvidia_cusolver_cu12-11.7.3.90-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:4376c11ad263152bd50ea295c05370360776f8c3427b30991df774f9fb26c450", size = 267506905, upload-time = "2025-03-07T01:47:16.273Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/bb/2e60de9bb1f0c3395eabd91ccad00f4ba3ef736dc9190a158a9d268419f5/nvidia_cusolver-12.0.3.29-py3-none-manylinux_2_27_aarch64.whl", hash = "sha256:3bb6e65ce0beaeafdd069b320246e8f17c1cd30ddb27a0539143a3706733a4d8", size = 193104180, upload-time = "2025-08-04T10:22:19.821Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/87/e3c9ee227b750e5b61572e7509f586cc8d494a4f7874b5163e734ed852c2/nvidia_cusolver-12.0.3.29-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:6f54c2eed5edab54c224dd1852dde80ba76b2b78e6d3ce7344fef5dfc66d16ab", size = 193474165, upload-time = "2025-08-04T10:22:47.976Z" },
 ]
 
 [[package]]
-name = "nvidia-cusparse-cu12"
+name = "nvidia-cusparse"
-version = "12.5.8.93"
+version = "12.6.2.49"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "nvidia-nvjitlink-cu12", marker = "(platform_machine != 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32')" },
+    { name = "nvidia-nvjitlink", marker = "sys_platform == 'linux'" },
 ]
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/c2/f5/e1854cb2f2bcd4280c44736c93550cc300ff4b8c95ebe370d0aa7d2b473d/nvidia_cusparse_cu12-12.5.8.93-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:1ec05d76bbbd8b61b06a80e1eaf8cf4959c3d4ce8e711b65ebd0443bb0ebb13b", size = 288216466, upload-time = "2025-03-07T01:48:13.779Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/30/f32023427f2ef4ec27e8293dfddb5068de566912cd0a45eccfd400017a62/nvidia_cusparse-12.6.2.49-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:5d3269c19283a0057fb5ebfb003ae2a10c97a28a6958f4238354826b055827c7", size = 155888587, upload-time = "2025-08-04T10:23:04.091Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/e8/b3f7a87cc719dca926c7baee92f2544de8909573a4126c85a9f1625431e8/nvidia_cusparse-12.6.2.49-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:efcf0b01e3a0827c144feff5391456b8a06e9ce63dcd51c0943e32e605251952", size = 140247612, upload-time = "2025-08-04T10:23:29.844Z" },
 ]
 
 [[package]]
-name = "nvidia-cusparselt-cu12"
+name = "nvidia-cusparselt-cu13"
-version = "0.7.1"
+version = "0.8.0"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/56/79/12978b96bd44274fe38b5dde5cfb660b1d114f70a65ef962bcbbed99b549/nvidia_cusparselt_cu12-0.7.1-py3-none-manylinux2014_x86_64.whl", hash = "sha256:f1bb701d6b930d5a7cea44c19ceb973311500847f81b634d802b7b539dc55623", size = 287193691, upload-time = "2025-02-26T00:15:44.104Z" },
+    { url = "https://files.pythonhosted.org/packages/46/10/8dcd1175260706a2fc92a16a52e306b71d4c1ea0b0cc4a9484183399818a/nvidia_cusparselt_cu13-0.8.0-py3-none-manylinux2014_aarch64.whl", hash = "sha256:400c6ed1cf6780fc6efedd64ec9f1345871767e6a1a0a552a1ea0578117ea77c", size = 220791277, upload-time = "2025-08-13T19:22:40.982Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/53/43b0d71f4e702fa9733f8b4571fdca50a8813f1e450b656c239beff12315/nvidia_cusparselt_cu13-0.8.0-py3-none-manylinux2014_x86_64.whl", hash = "sha256:25e30a8a7323935d4ad0340b95a0b69926eee755767e8e0b1cf8dd85b197d3fd", size = 169884119, upload-time = "2025-08-13T19:23:41.967Z" },
 ]
 
 [[package]]
-name = "nvidia-nccl-cu12"
+name = "nvidia-nccl-cu13"
-version = "2.27.5"
+version = "2.27.7"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/6e/89/f7a07dc961b60645dbbf42e80f2bc85ade7feb9a491b11a1e973aa00071f/nvidia_nccl_cu12-2.27.5-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:ad730cf15cb5d25fe849c6e6ca9eb5b76db16a80f13f425ac68d8e2e55624457", size = 322348229, upload-time = "2025-06-26T04:11:28.385Z" },
+    { url = "https://files.pythonhosted.org/packages/49/61/2c7762da6febee96341ea17d1f7309ac7559ac3cab00f3f7e1e7bd0e5d00/nvidia_nccl_cu13-2.27.7-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:5e3cc863e52bf9dd1e3ab1941bddb414098f489ae7342f6b3a274602303da123", size = 194014855, upload-time = "2025-09-23T16:30:27.56Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/3a/dabb10684e60edfaf1a1c9984d12a668bc1091582099d4e03ac5b9983b51/nvidia_nccl_cu13-2.27.7-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:b28a524abd8389b76a4a3f133c76a7aaa7005e47fcaa9d9603b90103927a3f93", size = 193901479, upload-time = "2025-09-23T16:30:41.165Z" },
 ]
 
 [[package]]
-name = "nvidia-nvjitlink-cu12"
+name = "nvidia-nvjitlink"
-version = "12.8.93"
+version = "13.0.39"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/f6/74/86a07f1d0f42998ca31312f998bd3b9a7eff7f52378f4f270c8679c77fb9/nvidia_nvjitlink_cu12-12.8.93-py3-none-manylinux2010_x86_64.manylinux_2_12_x86_64.whl", hash = "sha256:81ff63371a7ebd6e6451970684f916be2eab07321b73c9d244dc2b4da7f73b88", size = 39254836, upload-time = "2025-03-07T01:49:55.661Z" },
+    { url = "https://files.pythonhosted.org/packages/95/39/726edebeb76f3efc25c79f885429fa1227c9d200e20ea219bf724b382e19/nvidia_nvjitlink-13.0.39-py3-none-manylinux2010_x86_64.manylinux_2_12_x86_64.whl", hash = "sha256:bc3179be558329ef9687884c6faa27cdc0659bdbc642432ec8cc6cc00d182627", size = 40709605, upload-time = "2025-08-04T10:25:04.129Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/7a/0fb4c4413b3b14519f8934edd4dcd9f411c4e14e2a2c0ae58709e4dda255/nvidia_nvjitlink-13.0.39-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:ce0d63fa5ebedf542056e7491c49feed2297c900980aa6269b6a55f478056ad7", size = 38767126, upload-time = "2025-08-04T10:24:53.05Z" },
 ]
 
 [[package]]
-name = "nvidia-nvshmem-cu12"
+name = "nvidia-nvshmem-cu13"
-version = "3.3.20"
+version = "3.3.24"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/3b/6c/99acb2f9eb85c29fc6f3a7ac4dccfd992e22666dd08a642b303311326a97/nvidia_nvshmem_cu12-3.3.20-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:d00f26d3f9b2e3c3065be895e3059d6479ea5c638a3f38c9fec49b1b9dd7c1e5", size = 124657145, upload-time = "2025-08-04T20:25:19.995Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/7e/b8797780e442eabd9046cd6eb54100b8d0cb047ebc2f70931710cb03bcfe/nvidia_nvshmem_cu13-3.3.24-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:28ae82a4d14b322b93409535de62df6b7b83f4f7672ca97fc89107c2d40ce2c2", size = 60168129, upload-time = "2025-08-22T19:56:28.818Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/e9/8530afb8ed38d16bbc89cec80a4dd6a52dbf59bc93e546c3658cfa8b1f9b/nvidia_nvshmem_cu13-3.3.24-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:c14d09571697d2e57cb079c8daec88ab1c68cb3586532bfbd4886125a08339b7", size = 60390470, upload-time = "2025-08-22T19:56:49.848Z" },
 ]
 
 [[package]]
-name = "nvidia-nvtx-cu12"
+name = "nvidia-nvtx"
-version = "12.8.90"
+version = "13.0.39"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/a2/eb/86626c1bbc2edb86323022371c39aa48df6fd8b0a1647bc274577f72e90b/nvidia_nvtx_cu12-12.8.90-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:5b17e2001cc0d751a5bc2c6ec6d26ad95913324a4adb86788c944f8ce9ba441f", size = 89954, upload-time = "2025-03-07T01:42:44.131Z" },
+    { url = "https://files.pythonhosted.org/packages/38/37/0d103c84e7884382a79a569b720965141f83dd1c5df9e3e00cbc02d7099c/nvidia_nvtx-13.0.39-py3-none-manylinux1_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:cc113127785c96db8a0fe715df92db9788777b4b3d1bd713d42f75969201b5ce", size = 147197, upload-time = "2025-08-04T10:18:39.829Z" },
+    { url = "https://files.pythonhosted.org/packages/86/91/8b486ba85f71a2859dd705a4ec6aab38c37a389b8b7f94343db027732999/nvidia_nvtx-13.0.39-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:cddd2e08b35144f1000631c3880c9ebbcb8a2863d762e76f92d47d30ecaf87cc", size = 148037, upload-time = "2025-08-04T10:18:31.763Z" },
 ]
 
 [[package]]
@@ -677,40 +697,65 @@ wheels = [
 name = "torch"
 version = "2.9.1"
 source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "sys_platform == 'darwin'",
+    "sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
+]
 dependencies = [
-    { name = "filelock" },
+    { name = "filelock", marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
-    { name = "fsspec" },
+    { name = "fsspec", marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
-    { name = "jinja2" },
+    { name = "jinja2", marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
-    { name = "networkx" },
+    { name = "networkx", marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
-    { name = "nvidia-cublas-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "setuptools", marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
-    { name = "nvidia-cuda-cupti-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "sympy", marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
-    { name = "nvidia-cuda-nvrtc-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
+    { name = "typing-extensions", marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
-    { name = "nvidia-cuda-runtime-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-cudnn-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-cufft-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-cufile-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-curand-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-cusolver-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-cusparse-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-cusparselt-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-nccl-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-nvjitlink-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-nvshmem-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "nvidia-nvtx-cu12", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "setuptools" },
-    { name = "sympy" },
-    { name = "triton", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "typing-extensions" },
 ]
 wheels = [
     { url = "https://files.pythonhosted.org/packages/48/50/c4b5112546d0d13cc9eaa1c732b823d676a9f49ae8b6f97772f795874a03/torch-2.9.1-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:1edee27a7c9897f4e0b7c14cfc2f3008c571921134522d5b9b5ec4ebbc69041a", size = 74433245, upload-time = "2025-11-12T15:22:39.027Z" },
-    { url = "https://files.pythonhosted.org/packages/81/c9/2628f408f0518b3bae49c95f5af3728b6ab498c8624ab1e03a43dd53d650/torch-2.9.1-cp314-cp314-manylinux_2_28_aarch64.whl", hash = "sha256:19d144d6b3e29921f1fc70503e9f2fc572cde6a5115c0c0de2f7ca8b1483e8b6", size = 104134804, upload-time = "2025-11-12T15:22:35.222Z" },
-    { url = "https://files.pythonhosted.org/packages/28/fc/5bc91d6d831ae41bf6e9e6da6468f25330522e92347c9156eb3f1cb95956/torch-2.9.1-cp314-cp314-manylinux_2_28_x86_64.whl", hash = "sha256:c432d04376f6d9767a9852ea0def7b47a7bbc8e7af3b16ac9cf9ce02b12851c9", size = 899747132, upload-time = "2025-11-12T15:23:36.068Z" },
-    { url = "https://files.pythonhosted.org/packages/63/5d/e8d4e009e52b6b2cf1684bde2a6be157b96fb873732542fb2a9a99e85a83/torch-2.9.1-cp314-cp314-win_amd64.whl", hash = "sha256:d187566a2cdc726fc80138c3cdb260970fab1c27e99f85452721f7759bbd554d", size = 110934845, upload-time = "2025-11-12T15:22:48.367Z" },
     { url = "https://files.pythonhosted.org/packages/bd/b2/2d15a52516b2ea3f414643b8de68fa4cb220d3877ac8b1028c83dc8ca1c4/torch-2.9.1-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:cb10896a1f7fedaddbccc2017ce6ca9ecaaf990f0973bdfcf405439750118d2c", size = 74823558, upload-time = "2025-11-12T15:22:43.392Z" },
-    { url = "https://files.pythonhosted.org/packages/86/5c/5b2e5d84f5b9850cd1e71af07524d8cbb74cba19379800f1f9f7c997fc70/torch-2.9.1-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:0a2bd769944991c74acf0c4ef23603b9c777fdf7637f115605a4b2d8023110c7", size = 104145788, upload-time = "2025-11-12T15:23:52.109Z" },
+]
-    { url = "https://files.pythonhosted.org/packages/a9/8c/3da60787bcf70add986c4ad485993026ac0ca74f2fc21410bc4eb1bb7695/torch-2.9.1-cp314-cp314t-manylinux_2_28_x86_64.whl", hash = "sha256:07c8a9660bc9414c39cac530ac83b1fb1b679d7155824144a40a54f4a47bfa73", size = 899735500, upload-time = "2025-11-12T15:24:08.788Z" },
+
-    { url = "https://files.pythonhosted.org/packages/db/2b/f7818f6ec88758dfd21da46b6cd46af9d1b3433e53ddbb19ad1e0da17f9b/torch-2.9.1-cp314-cp314t-win_amd64.whl", hash = "sha256:c88d3299ddeb2b35dcc31753305612db485ab6f1823e37fb29451c8b2732b87e", size = 111163659, upload-time = "2025-11-12T15:23:20.009Z" },
+[[package]]
+name = "torch"
+version = "2.9.1+cu130"
+source = { registry = "https://download.pytorch.org/whl/cu130" }
+resolution-markers = [
+    "sys_platform == 'win32'",
+    "platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "platform_machine != 'aarch64' and sys_platform == 'linux'",
+]
+dependencies = [
+    { name = "filelock", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+    { name = "fsspec", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+    { name = "jinja2", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+    { name = "networkx", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+    { name = "nvidia-cublas", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cuda-cupti", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cuda-nvrtc", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cuda-runtime", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cudnn-cu13", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cufft", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cufile", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-curand", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cusolver", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cusparse", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cusparselt-cu13", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-nccl-cu13", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-nvjitlink", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-nvshmem-cu13", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-nvtx", marker = "sys_platform == 'linux'" },
+    { name = "setuptools", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+    { name = "sympy", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+    { name = "triton", marker = "sys_platform == 'linux'" },
+    { name = "typing-extensions", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
+]
+wheels = [
+    { url = "https://download.pytorch.org/whl/cu130/torch-2.9.1%2Bcu130-cp314-cp314-manylinux_2_28_aarch64.whl" },
+    { url = "https://download.pytorch.org/whl/cu130/torch-2.9.1%2Bcu130-cp314-cp314-manylinux_2_28_x86_64.whl" },
+    { url = "https://download.pytorch.org/whl/cu130/torch-2.9.1%2Bcu130-cp314-cp314-win_amd64.whl" },
+    { url = "https://download.pytorch.org/whl/cu130/torch-2.9.1%2Bcu130-cp314-cp314t-manylinux_2_28_aarch64.whl" },
+    { url = "https://download.pytorch.org/whl/cu130/torch-2.9.1%2Bcu130-cp314-cp314t-manylinux_2_28_x86_64.whl" },
+    { url = "https://download.pytorch.org/whl/cu130/torch-2.9.1%2Bcu130-cp314-cp314t-win_amd64.whl" },
 ]
 
 [[package]]
@@ -720,7 +765,8 @@ source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "numpy" },
     { name = "pillow" },
-    { name = "torch" },
+    { name = "torch", version = "2.9.1", source = { registry = "https://pypi.org/simple" }, marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
+    { name = "torch", version = "2.9.1+cu130", source = { registry = "https://download.pytorch.org/whl/cu130" }, marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
 ]
 wheels = [
     { url = "https://files.pythonhosted.org/packages/42/84/577b2cef8f32094add5f52887867da4c2a3e6b4261538447e9b48eb25812/torchvision-0.24.1-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:cccf4b4fec7fdfcd3431b9ea75d1588c0a8596d0333245dafebee0462abe3388", size = 2005319, upload-time = "2025-11-12T15:25:23.827Z" },
@@ -738,7 +784,9 @@ name = "triton"
 version = "3.5.1"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
+    { url = "https://files.pythonhosted.org/packages/14/f9/0430e879c1e63a1016cb843261528fd3187c872c3a9539132efc39514753/triton-3.5.1-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:f617aa7925f9ea9968ec2e1adaf93e87864ff51549c8f04ce658f29bbdb71e2d", size = 159956163, upload-time = "2025-11-11T17:52:12.999Z" },
     { url = "https://files.pythonhosted.org/packages/a4/e6/c595c35e5c50c4bc56a7bac96493dad321e9e29b953b526bbbe20f9911d0/triton-3.5.1-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d0637b1efb1db599a8e9dc960d53ab6e4637db7d4ab6630a0974705d77b14b60", size = 170480488, upload-time = "2025-11-11T17:41:18.222Z" },
+    { url = "https://files.pythonhosted.org/packages/41/1e/63d367c576c75919e268e4fbc33c1cb33b6dc12bb85e8bfe531c2a8bd5d3/triton-3.5.1-cp314-cp314t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:8932391d7f93698dfe5bc9bead77c47a24f97329e9f20c10786bb230a9083f56", size = 160073620, upload-time = "2025-11-11T17:52:18.403Z" },
     { url = "https://files.pythonhosted.org/packages/16/b5/b0d3d8b901b6a04ca38df5e24c27e53afb15b93624d7fd7d658c7cd9352a/triton-3.5.1-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:bac7f7d959ad0f48c0e97d6643a1cc0fd5786fe61cb1f83b537c6b2d54776478", size = 170582192, upload-time = "2025-11-11T17:41:23.963Z" },
 ]
 
@@ -765,7 +813,8 @@ dependencies = [
     { name = "pyyaml" },
     { name = "requests" },
     { name = "scipy" },
-    { name = "torch" },
+    { name = "torch", version = "2.9.1", source = { registry = "https://pypi.org/simple" }, marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
+    { name = "torch", version = "2.9.1+cu130", source = { registry = "https://download.pytorch.org/whl/cu130" }, marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
     { name = "torchvision" },
     { name = "ultralytics-thop" },
 ]
@@ -780,7 +829,8 @@ version = "2.0.18"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "numpy" },
-    { name = "torch" },
+    { name = "torch", version = "2.9.1", source = { registry = "https://pypi.org/simple" }, marker = "sys_platform != 'linux' and sys_platform != 'win32'" },
+    { name = "torch", version = "2.9.1+cu130", source = { registry = "https://download.pytorch.org/whl/cu130" }, marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/1f/63/21a32e1facfeee245dbdfb7b4669faf7a36ff7c00b50987932bdab126f4b/ultralytics_thop-2.0.18.tar.gz", hash = "sha256:21103bcd39cc9928477dc3d9374561749b66a1781b35f46256c8d8c4ac01d9cf", size = 34557, upload-time = "2025-10-29T16:58:13.526Z" }
 wheels = [