Make dora cuda agnostic and compatible using zig

.github/workflows/node-hub-ci-cd.yml

@@ -141,7 +141,7 @@ jobs:
             if [[ -f "Cargo.toml" && -f "pyproject.toml" ]]; then
               echo "Publishing $dir using maturin..."
               pip3 install "maturin[patchelf]"                                                                             
-              maturin publish  --skip-existing
+              maturin publish  --skip-existing --zig
             else
               if [ -f "pyproject.toml" ]; then
                 echo "Publishing $dir using Poetry..."
@@ -156,7 +156,7 @@ jobs:
                     echo "Package '$package_name' version '$version' already exists on crates.io. Skipping publish."
                 else
                     echo "Publishing package '$package_name' version '$version'..."
-                    cargo publish
+                    cargo publish 
                 fi
               fi
             fi

node-hub/dora-internvl/dora_internvl/main.py

@@ -101,6 +101,7 @@ def load_image(image_array: np.array, input_size=448, max_num=12):
 def main():
     # Handle dynamic nodes, ask for the name of the node in the dataflow, and the same values as the ENV variables.
     model_path = os.getenv("MODEL", "OpenGVLab/InternVL2-1B")
+    device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
     # If you want to load a model using multiple GPUs, please refer to the `Multiple GPUs` section.
     model = (
@@ -112,7 +113,7 @@ def main():
             trust_remote_code=True,
         )
         .eval()
-        .cuda()
+        .to(device)
     )
     tokenizer = AutoTokenizer.from_pretrained(
         model_path, trust_remote_code=True, use_fast=False

node-hub/dora-qwenvl/dora_qwenvl/main.py

@@ -1,5 +1,6 @@
 import os
 from dora import Node
+import torch
 from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
 from qwen_vl_utils import process_vision_info
 import numpy as np
@@ -83,7 +84,9 @@ def generate(frames: dict, question):
         padding=True,
         return_tensors="pt",
     )
-    inputs = inputs.to("cuda")
+
+    device = "cuda:0" if torch.cuda.is_available() else "cpu"
+    inputs = inputs.to(device)
 
     # Inference: Generation of the output
     generated_ids = model.generate(**inputs, max_new_tokens=128)

node-hub/dora-qwenvl/pyproject.toml

@@ -14,7 +14,7 @@ packages = [{ include = "dora_qwenvl" }]
 python = "^3.7"
 dora-rs = "^0.3.6"
 numpy = "< 2.0.0"
-torch = "^2.4.0"
+torch = "^2.2.0"
 torchvision = "^0.19"
 transformers = "^4.45"
 qwen-vl-utils = "^0.0.2"