remove comments

Author: raphaelamorim

run-nemotron-v2-VL/build.sh

@@ -3,23 +3,11 @@
 
 # Launch Docker container with NVIDIA GPU support
 docker run --runtime nvidia --gpus all \
-      # Increase memory lock and stack size limits for GPU operations
       --ulimit memlock=-1 --ulimit stack=67108864 \
-      # Mount HuggingFace cache to avoid re-downloading models
       -v ~/.cache/huggingface:/root/.cache/huggingface \
-      # Mount vLLM cache for compiled kernels and other artifacts
       -v ~/.cache/vllm:/root/.cache/vllm \
-      # Expose port 8000 for vLLM API server
       -p 8000:8000 \
-      # Set HuggingFace token for model access
       --env "HUGGING_FACE_HUB_TOKEN=<YOUR_HUGGINGFACE_TOKEN>" \
-      # Set Triton PTXAS compiler path for CUDA compilation
       --env "TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas" \
-      # Use the vllm:25.10 image built by build.sh
       vllm:25.10 \
-      # Serve Nemotron Nano VL 12B model with FP4 quantization
-      # --trust-remote-code: Allow execution of custom model code
-      # --quantization modelopt_fp4: Use FP4 quantization for reduced memory
-      # --max-model-len 24000: Set maximum sequence length
-      # --gpu-memory-utilization 0.3: Use 30% of GPU memory for model weights
       vllm serve nvidia/Nemotron-Nano-VL-12B-V2-FP4-QAD --trust-remote-code --quantization modelopt_fp4 --max-model-len 24000 --gpu-memory-utilization 0.3