mirror of
https://github.com/NVIDIA/dgx-spark-playbooks.git
synced 2026-04-22 01:53:53 +00:00
chore: Regenerate all playbooks
This commit is contained in:
GitLab CI
parent
895e2d9d69
commit
6818481902
@ -5,36 +5,20 @@
|
||||
## Table of Contents
|
||||
|
||||
- [Overview](#overview)
|
||||
- [What you'll accomplish](#what-youll-accomplish)
|
||||
- [What to know before starting](#what-to-know-before-starting)
|
||||
- [Prerequisites](#prerequisites)
|
||||
- [Ancillary files](#ancillary-files)
|
||||
- [Time & risk](#time-risk)
|
||||
- [Instructions](#instructions)
|
||||
- [Step 1. Verify system prerequisites](#step-1-verify-system-prerequisites)
|
||||
- [Step 2. Launch PyTorch container with GPU support](#step-2-launch-pytorch-container-with-gpu-support)
|
||||
- [Step 3. Clone LLaMA Factory repository](#step-3-clone-llama-factory-repository)
|
||||
- [Step 4. Install LLaMA Factory with dependencies](#step-4-install-llama-factory-with-dependencies)
|
||||
- [Step 5. Configure PyTorch for CUDA 12.9 (if needed)](#step-5-configure-pytorch-for-cuda-129-if-needed)
|
||||
- [Step 6. Prepare training configuration](#step-6-prepare-training-configuration)
|
||||
- [Step 7. Launch fine-tuning training](#step-7-launch-fine-tuning-training)
|
||||
- [Step 8. Validate training completion](#step-8-validate-training-completion)
|
||||
- [Step 9. Test inference with fine-tuned model](#step-9-test-inference-with-fine-tuned-model)
|
||||
- [Step 10. Troubleshooting](#step-10-troubleshooting)
|
||||
- [Step 11. Cleanup and rollback](#step-11-cleanup-and-rollback)
|
||||
- [Step 12. Next steps](#step-12-next-steps)
|
||||
|
||||
---
|
||||
|
||||
## Overview
|
||||
|
||||
### What you'll accomplish
|
||||
## What you'll accomplish
|
||||
|
||||
You'll set up LLaMA Factory on NVIDIA Spark with Blackwell architecture to fine-tune large
|
||||
language models using LoRA, QLoRA, and full fine-tuning methods. This enables efficient
|
||||
model adaptation for specialized domains while leveraging hardware-specific optimizations.
|
||||
|
||||
### What to know before starting
|
||||
## What to know before starting
|
||||
|
||||
- Basic Python knowledge for editing config files and troubleshooting
|
||||
- Command line usage for running shell commands and managing environments
|
||||
@ -44,7 +28,7 @@ model adaptation for specialized domains while leveraging hardware-specific opti
|
||||
- Dataset preparation: formatting text data into JSON structure for instruction tuning
|
||||
- Resource management: adjusting batch size and memory settings for GPU constraints
|
||||
|
||||
### Prerequisites
|
||||
## Prerequisites
|
||||
|
||||
- NVIDIA Spark device with Blackwell architecture
|
||||
|
||||
@ -60,7 +44,7 @@ model adaptation for specialized domains while leveraging hardware-specific opti
|
||||
|
||||
- Internet connection for downloading models from Hugging Face Hub
|
||||
|
||||
### Ancillary files
|
||||
## Ancillary files
|
||||
|
||||
- Official LLaMA Factory repository: https://github.com/hiyouga/LLaMA-Factory
|
||||
|
||||
@ -70,7 +54,7 @@ model adaptation for specialized domains while leveraging hardware-specific opti
|
||||
|
||||
- Documentation: https://llamafactory.readthedocs.io/en/latest/getting_started/data_preparation.html
|
||||
|
||||
### Time & risk
|
||||
## Time & risk
|
||||
|
||||
**Duration:** 30-60 minutes for initial setup, 1-7 hours for training depending on model size
|
||||
and dataset.
|
||||
@ -83,7 +67,7 @@ saved locally and can be deleted to reclaim storage space.
|
||||
|
||||
## Instructions
|
||||
|
||||
### Step 1. Verify system prerequisites
|
||||
## Step 1. Verify system prerequisites
|
||||
|
||||
Check that your NVIDIA Spark system has the required components installed and accessible.
|
||||
|
||||
@ -95,7 +79,7 @@ python --version
|
||||
git --version
|
||||
```
|
||||
|
||||
### Step 2. Launch PyTorch container with GPU support
|
||||
## Step 2. Launch PyTorch container with GPU support
|
||||
|
||||
Start the NVIDIA PyTorch container with GPU access and mount your workspace directory.
|
||||
> **Note:** This NVIDIA PyTorch container supports CUDA 13
|
||||
@ -104,7 +88,7 @@ Start the NVIDIA PyTorch container with GPU access and mount your workspace dire
|
||||
docker run --gpus all --ipc=host --ulimit memlock=-1 -it --ulimit stack=67108864 --rm -v "$PWD":/workspace nvcr.io/nvidia/pytorch:25.08-py3 bash
|
||||
```
|
||||
|
||||
### Step 3. Clone LLaMA Factory repository
|
||||
## Step 3. Clone LLaMA Factory repository
|
||||
|
||||
Download the LLaMA Factory source code from the official repository.
|
||||
|
||||
@ -121,9 +105,9 @@ Install the package in editable mode with metrics support for training evaluatio
|
||||
pip install -e ".[metrics]"
|
||||
```
|
||||
|
||||
### Step 5. Configure PyTorch for CUDA 12.9 (if needed)
|
||||
## Step 5. Configure PyTorch for CUDA 12.9 (if needed)
|
||||
|
||||
#### If using standalone Python (skip if using Docker container)
|
||||
*If using standalone Python (skip if using Docker container)*
|
||||
|
||||
In a python virtual environment, uninstall existing PyTorch and reinstall with CUDA 12.9 support for ARM64 architecture.
|
||||
|
||||
@ -132,7 +116,7 @@ pip uninstall torch torchvision torchaudio
|
||||
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu129
|
||||
```
|
||||
|
||||
#### If using Docker container
|
||||
*If using Docker container*
|
||||
|
||||
PyTorch is pre-installed with CUDA support. Verify installation:
|
||||
|
||||
@ -140,7 +124,7 @@ PyTorch is pre-installed with CUDA support. Verify installation:
|
||||
python -c "import torch; print(f'PyTorch: {torch.__version__}, CUDA: {torch.cuda.is_available()}')"
|
||||
```
|
||||
|
||||
### Step 6. Prepare training configuration
|
||||
## Step 6. Prepare training configuration
|
||||
|
||||
Examine the provided LoRA fine-tuning configuration for Llama-3.
|
||||
|
||||
@ -148,7 +132,7 @@ Examine the provided LoRA fine-tuning configuration for Llama-3.
|
||||
cat examples/train_lora/llama3_lora_sft.yaml
|
||||
```
|
||||
|
||||
### Step 7. Launch fine-tuning training
|
||||
## Step 7. Launch fine-tuning training
|
||||
|
||||
> **Note:** Login to your hugging face hub to download the model if the model is gated
|
||||
Execute the training process using the pre-configured LoRA setup.
|
||||
@ -170,7 +154,7 @@ Example output:
|
||||
Figure saved at: saves/llama3-8b/lora/sft/training_loss.png
|
||||
```
|
||||
|
||||
### Step 8. Validate training completion
|
||||
## Step 8. Validate training completion
|
||||
|
||||
Verify that training completed successfully and checkpoints were saved.
|
||||
|
||||
@ -186,7 +170,7 @@ Expected output should show:
|
||||
- Training metrics showing decreasing loss values
|
||||
- Training loss plot saved as PNG file
|
||||
|
||||
### Step 9. Test inference with fine-tuned model
|
||||
## Step 9. Test inference with fine-tuned model
|
||||
|
||||
Run a simple inference test to verify the fine-tuned model loads correctly.
|
||||
|
||||
@ -194,7 +178,7 @@ Run a simple inference test to verify the fine-tuned model loads correctly.
|
||||
llamafactory-cli chat examples/inference/llama3_lora_sft.yaml
|
||||
```
|
||||
|
||||
### Step 10. Troubleshooting
|
||||
## Step 10. Troubleshooting
|
||||
|
||||
| Symptom | Cause | Fix |
|
||||
|---------|--------|-----|
|
||||
@ -202,7 +186,7 @@ llamafactory-cli chat examples/inference/llama3_lora_sft.yaml
|
||||
| Model download fails or is slow | Network connectivity or Hugging Face Hub issues | Check internet connection, try using `HF_HUB_OFFLINE=1` for cached models |
|
||||
| Training loss not decreasing | Learning rate too high/low or insufficient data | Adjust `learning_rate` parameter or check dataset quality |
|
||||
|
||||
### Step 11. Cleanup and rollback
|
||||
## Step 11. Cleanup and rollback
|
||||
|
||||
> **Warning:** This will delete all training progress and checkpoints.
|
||||
|
||||
@ -220,7 +204,7 @@ exit # Exit container
|
||||
docker container prune -f
|
||||
```
|
||||
|
||||
### Step 12. Next steps
|
||||
## Step 12. Next steps
|
||||
|
||||
Test your fine-tuned model with custom prompts:
|
||||
|
||||
|
||||
@ -35,14 +35,14 @@ FP8, FP4).
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA Spark device with Blackwell GPU architecture
|
||||
- [ ] Docker installed and accessible to current user
|
||||
- [ ] NVIDIA Container Runtime configured
|
||||
- [ ] Hugging Face account with valid token
|
||||
- [ ] At least 48GB VRAM available for FP16 Flux.1 Schnell operations
|
||||
- [ ] Verify GPU access: `nvidia-smi`
|
||||
- [ ] Check Docker GPU integration: `docker run --rm --gpus all nvidia/cuda:12.0-base-ubuntu20.04 nvidia-smi`
|
||||
- [ ] Confirm HF token access with permissions to FLUX repos: `echo $HF_TOKEN`, Sign in to your huggingface account You can create the token from create your token here (make sure you provide permissions to the token): https://huggingface.co/settings/tokens , Note the permissions to be checked and the repos: black-forest-labs/FLUX.1-dev and black-forest-labs/FLUX.1-dev-onnx (search for these repos when creating the user token) to be added.
|
||||
- NVIDIA Spark device with Blackwell GPU architecture
|
||||
- Docker installed and accessible to current user
|
||||
- NVIDIA Container Runtime configured
|
||||
- Hugging Face account with valid token
|
||||
- At least 48GB VRAM available for FP16 Flux.1 Schnell operations
|
||||
- Verify GPU access: `nvidia-smi`
|
||||
- Check Docker GPU integration: `docker run --rm --gpus all nvidia/cuda:12.0-base-ubuntu20.04 nvidia-smi`
|
||||
- Confirm HF token access with permissions to FLUX repos: `echo $HF_TOKEN`, Sign in to your huggingface account You can create the token from create your token here (make sure you provide permissions to the token): https://huggingface.co/settings/tokens , Note the permissions to be checked and the repos: black-forest-labs/FLUX.1-dev and black-forest-labs/FLUX.1-dev-onnx (search for these repos when creating the user token) to be added.
|
||||
|
||||
## Ancillary files
|
||||
|
||||
|
||||
@ -35,22 +35,22 @@ You'll establish a complete fine-tuning environment for large language models (1
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA Spark device with Blackwell architecture GPU access
|
||||
- [ ] CUDA toolkit 12.0+ installed and configured
|
||||
- NVIDIA Spark device with Blackwell architecture GPU access
|
||||
- CUDA toolkit 12.0+ installed and configured
|
||||
```bash
|
||||
nvcc --version
|
||||
```
|
||||
- [ ] Python 3.10+ environment available
|
||||
- Python 3.10+ environment available
|
||||
```bash
|
||||
python3 --version
|
||||
```
|
||||
- [ ] Minimum 32GB system RAM for efficient model loading and training
|
||||
- [ ] Active internet connection for downloading models and packages
|
||||
- [ ] Git installed for repository cloning
|
||||
- Minimum 32GB system RAM for efficient model loading and training
|
||||
- Active internet connection for downloading models and packages
|
||||
- Git installed for repository cloning
|
||||
```bash
|
||||
git --version
|
||||
```
|
||||
- [ ] SSH access to your NVIDIA Spark device configured
|
||||
- SSH access to your NVIDIA Spark device configured
|
||||
|
||||
## Ancillary files
|
||||
|
||||
|
||||
@ -40,19 +40,19 @@ completions.
|
||||
|
||||
### Prerequisites
|
||||
|
||||
- [ ] DGX Spark device with NVIDIA drivers installed
|
||||
- DGX Spark device with NVIDIA drivers installed
|
||||
```bash
|
||||
nvidia-smi
|
||||
```
|
||||
- [ ] Docker with NVIDIA Container Toolkit configured, instructions here: https://******.nvidia.com/dgx-docs/review/621/dgx-spark/latest/nvidia-container-runtime-for-docker.html
|
||||
- Docker with NVIDIA Container Toolkit configured, instructions here: https://******.nvidia.com/dgx-docs/review/621/dgx-spark/latest/nvidia-container-runtime-for-docker.html
|
||||
```bash
|
||||
docker run -it --gpus=all nvcr.io/nvidia/cuda:13.0.1-devel-ubuntu24.04 nvidia-smi
|
||||
```
|
||||
- [ ] NGC account with API key from https://ngc.nvidia.com/setup/api-key
|
||||
- NGC account with API key from https://ngc.nvidia.com/setup/api-key
|
||||
```bash
|
||||
echo $NGC_API_KEY | grep -E '^[a-zA-Z0-9]{86}=='
|
||||
```
|
||||
- [ ] Sufficient disk space for model caching (varies by model, typically 10-50GB)
|
||||
- Sufficient disk space for model caching (varies by model, typically 10-50GB)
|
||||
```bash
|
||||
df -h ~
|
||||
```
|
||||
|
||||
@ -40,11 +40,11 @@ inside a TensorRT-LLM container, producing an NVFP4 quantized model for deployme
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA Spark device with Blackwell architecture GPU
|
||||
- [ ] Docker installed with GPU support
|
||||
- [ ] NVIDIA Container Toolkit configured
|
||||
- [ ] At least 32GB of available storage for model files and outputs
|
||||
- [ ] Hugging Face account with access to the target model
|
||||
- NVIDIA Spark device with Blackwell architecture GPU
|
||||
- Docker installed with GPU support
|
||||
- NVIDIA Container Toolkit configured
|
||||
- At least 32GB of available storage for model files and outputs
|
||||
- Hugging Face account with access to the target model
|
||||
|
||||
Verify your setup:
|
||||
```bash
|
||||
|
||||
@ -36,12 +36,9 @@ the powerful GPU capabilities of your Spark device without complex network confi
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] DGX Spark device set up and connected to your network
|
||||
- Verify with: `nvidia-smi` (should show Blackwell GPU information)
|
||||
- [ ] NVIDIA Sync installed and connected to your Spark
|
||||
- Verify connection status in NVIDIA Sync system tray application
|
||||
- [ ] Terminal access to your local machine for testing API calls
|
||||
- Verify with: `curl --version`
|
||||
- DGX Spark device set up and connected to your network
|
||||
- NVIDIA Sync installed and connected to your Spark
|
||||
- Terminal access to your local machine for testing API calls
|
||||
|
||||
|
||||
|
||||
|
||||
@ -30,23 +30,23 @@ RTX Pro 6000 or DGX Spark workstation.
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA GPU (RTX Pro 6000 or DGX Spark recommended)
|
||||
- NVIDIA GPU (RTX Pro 6000 or DGX Spark recommended)
|
||||
```bash
|
||||
nvidia-smi # Should show GPU with CUDA ≥12.9
|
||||
```
|
||||
- [ ] NVIDIA drivers and CUDA toolkit installed
|
||||
- NVIDIA drivers and CUDA toolkit installed
|
||||
```bash
|
||||
nvcc --version # Should show CUDA 12.9 or higher
|
||||
```
|
||||
- [ ] Docker with NVIDIA Container Toolkit
|
||||
- Docker with NVIDIA Container Toolkit
|
||||
```bash
|
||||
docker run --rm --gpus all nvidia/cuda:12.9.0-base-ubuntu22.04 nvidia-smi
|
||||
```
|
||||
- [ ] Python 3.8+ environment
|
||||
- Python 3.8+ environment
|
||||
```bash
|
||||
python3 --version # Should show 3.8 or higher
|
||||
```
|
||||
- [ ] Sufficient disk space for databases (>3TB recommended)
|
||||
- Sufficient disk space for databases (>3TB recommended)
|
||||
```bash
|
||||
df -h # Check available space
|
||||
```
|
||||
|
||||
@ -35,12 +35,12 @@ vision-language tasks using models like DeepSeek-V2-Lite.
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA Spark device with Blackwell architecture
|
||||
- [ ] Docker Engine installed and running: `docker --version`
|
||||
- [ ] NVIDIA GPU drivers installed: `nvidia-smi`
|
||||
- [ ] NVIDIA Container Toolkit configured: `docker run --rm --gpus all nvidia/cuda:12.9-base nvidia-smi`
|
||||
- [ ] Sufficient disk space (>20GB available): `df -h`
|
||||
- [ ] Network connectivity for pulling NGC containers: `ping nvcr.io`
|
||||
- NVIDIA Spark device with Blackwell architecture
|
||||
- Docker Engine installed and running: `docker --version`
|
||||
- NVIDIA GPU drivers installed: `nvidia-smi`
|
||||
- NVIDIA Container Toolkit configured: `docker run --rm --gpus all nvidia/cuda:12.9-base nvidia-smi`
|
||||
- Sufficient disk space (>20GB available): `df -h`
|
||||
- Network connectivity for pulling NGC containers: `ping nvcr.io`
|
||||
|
||||
## Ancillary files
|
||||
|
||||
|
||||
@ -40,17 +40,17 @@ These examples demonstrate how to accelerate large language model inference whil
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA Spark device with sufficient GPU memory available (80GB+ recommended for GPT-OSS 120B)
|
||||
- [ ] Docker with GPU support enabled
|
||||
- NVIDIA Spark device with sufficient GPU memory available (80GB+ recommended for GPT-OSS 120B)
|
||||
- Docker with GPU support enabled
|
||||
```bash
|
||||
docker run --gpus all nvcr.io/nvidia/tensorrt-llm/release:spark-single-gpu-dev nvidia-smi
|
||||
```
|
||||
- [ ] Access to NVIDIA's internal container registry (for Eagle3 example)
|
||||
- [ ] HuggingFace authentication configured (if needed for model downloads)
|
||||
- Access to NVIDIA's internal container registry (for Eagle3 example)
|
||||
- HuggingFace authentication configured (if needed for model downloads)
|
||||
```bash
|
||||
huggingface-cli login
|
||||
```
|
||||
- [ ] Network connectivity for model downloads
|
||||
- Network connectivity for model downloads
|
||||
|
||||
|
||||
## Time & risk
|
||||
|
||||
@ -51,13 +51,13 @@ all traffic automatically encrypted and NAT traversal handled transparently.
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA Spark device running Ubuntu (ARM64/AArch64)
|
||||
- [ ] Client device (Mac, Windows, or Linux) for remote access
|
||||
- [ ] Internet connectivity on both devices
|
||||
- [ ] Valid email account for Tailscale authentication (Google, GitHub, Microsoft)
|
||||
- [ ] SSH server availability check: `systemctl status ssh`
|
||||
- [ ] Package manager working: `sudo apt update`
|
||||
- [ ] User account with sudo privileges on Spark device
|
||||
- NVIDIA Spark device running Ubuntu (ARM64/AArch64)
|
||||
- Client device (Mac, Windows, or Linux) for remote access
|
||||
- Internet connectivity on both devices
|
||||
- Valid email account for Tailscale authentication (Google, GitHub, Microsoft)
|
||||
- SSH server availability check: `systemctl status ssh`
|
||||
- Package manager working: `sudo apt update`
|
||||
- User account with sudo privileges on Spark device
|
||||
|
||||
## Time & risk
|
||||
|
||||
|
||||
@ -54,13 +54,13 @@ inference through kernel-level optimizations, efficient memory layouts, and adva
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA Spark device with Blackwell architecture GPUs
|
||||
- [ ] NVIDIA drivers compatible with CUDA 12.x: `nvidia-smi`
|
||||
- [ ] Docker installed and GPU support configured: `docker run --rm --gpus all nvcr.io/nvidia/tensorrt-llm/release:spark-single-gpu-dev nvidia-smi`
|
||||
- [ ] Hugging Face account with token for model access: `echo $HF_TOKEN`
|
||||
- [ ] Sufficient GPU VRAM (16GB+ recommended for 70B models)
|
||||
- [ ] Internet connectivity for downloading models and container images
|
||||
- [ ] Network: open TCP ports 8355 (LLM) and 8356 (VLM) on host for OpenAI-compatible serving
|
||||
- NVIDIA Spark device with Blackwell architecture GPUs
|
||||
- NVIDIA drivers compatible with CUDA 12.x: `nvidia-smi`
|
||||
- Docker installed and GPU support configured: `docker run --rm --gpus all nvcr.io/nvidia/tensorrt-llm/release:spark-single-gpu-dev nvidia-smi`
|
||||
- Hugging Face account with token for model access: `echo $HF_TOKEN`
|
||||
- Sufficient GPU VRAM (16GB+ recommended for 70B models)
|
||||
- Internet connectivity for downloading models and container images
|
||||
- Network: open TCP ports 8355 (LLM) and 8356 (VLM) on host for OpenAI-compatible serving
|
||||
|
||||
## Model Support Matrix
|
||||
|
||||
|
||||
@ -36,10 +36,10 @@ parameter-efficient fine-tuning methods like LoRA and QLoRA.
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA Spark device with Blackwell GPU architecture
|
||||
- [ ] `nvidia-smi` shows a summary of GPU information
|
||||
- [ ] CUDA 13.0 installed: `nvcc --version`
|
||||
- [ ] Internet access for downloading models and datasets
|
||||
- NVIDIA Spark device with Blackwell GPU architecture
|
||||
- `nvidia-smi` shows a summary of GPU information
|
||||
- CUDA 13.0 installed: `nvcc --version`
|
||||
- Internet access for downloading models and datasets
|
||||
|
||||
##Ancillary files
|
||||
|
||||
|
||||
@ -5,9 +5,9 @@
|
||||
## Table of Contents
|
||||
|
||||
- [Overview](#overview)
|
||||
- [Instructions](#instructions)
|
||||
- [Run on two Sparks](#run-on-two-sparks)
|
||||
- [Step 14. (Optional) Launch 405B inference server](#step-14-optional-launch-405b-inference-server)
|
||||
- [Access through terminal](#access-through-terminal)
|
||||
|
||||
---
|
||||
|
||||
@ -29,14 +29,14 @@ support for ARM64.
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] DGX Spark device with ARM64 processor and Blackwell GPU architecture
|
||||
- [ ] CUDA 12.9 or CUDA 13.0 toolkit installed: `nvcc --version` shows CUDA toolkit version.
|
||||
- [ ] Docker installed and configured: `docker --version` succeeds
|
||||
- [ ] NVIDIA Container Toolkit installed
|
||||
- [ ] Python 3.12 available: `python3.12 --version` succeeds
|
||||
- [ ] Git installed: `git --version` succeeds
|
||||
- [ ] Network access to download packages and container images
|
||||
- [ ] > TODO: Verify memory and storage requirements for builds
|
||||
- DGX Spark device with ARM64 processor and Blackwell GPU architecture
|
||||
- CUDA 12.9 or CUDA 13.0 toolkit installed: `nvcc --version` shows CUDA toolkit version.
|
||||
- Docker installed and configured: `docker --version` succeeds
|
||||
- NVIDIA Container Toolkit installed
|
||||
- Python 3.12 available: `python3.12 --version` succeeds
|
||||
- Git installed: `git --version` succeeds
|
||||
- Network access to download packages and container images
|
||||
|
||||
|
||||
## Time & risk
|
||||
|
||||
@ -46,6 +46,77 @@ support for ARM64.
|
||||
|
||||
**Rollback:** Container approach is non-destructive.
|
||||
|
||||
## Instructions
|
||||
|
||||
## Step 1. Pull vLLM container image
|
||||
|
||||
Find the latest container build from https://catalog.ngc.nvidia.com/orgs/nvidia/containers/vllm?version=25.09-py3
|
||||
```
|
||||
docker pull nvcr.io/nvidia/vllm:25.09-py3
|
||||
```
|
||||
|
||||
## Step 2. Test vLLM in container
|
||||
|
||||
Launch the container and start vLLM server with a test model to verify basic functionality.
|
||||
|
||||
```bash
|
||||
docker run -it --gpus all -p 8000:8000 \
|
||||
nvcr.io/nvidia/vllm:25.09-py3 \
|
||||
vllm serve "Qwen/Qwen2.5-Math-1.5B-Instruct"
|
||||
```
|
||||
|
||||
Expected output should include:
|
||||
- Model loading confirmation
|
||||
- Server startup on port 8000
|
||||
- GPU memory allocation details
|
||||
|
||||
In another terminal, test the server:
|
||||
|
||||
```bash
|
||||
curl http://localhost:8000/v1/chat/completions \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"model": "Qwen/Qwen2.5-Math-1.5B-Instruct",
|
||||
"messages": [{"role": "user", "content": "12*17"}],
|
||||
"max_tokens": 500
|
||||
}'
|
||||
```
|
||||
|
||||
Expected response should contain `"content": "204"` or similar mathematical calculation.
|
||||
|
||||
## Step 3. Troubleshooting
|
||||
|
||||
| Symptom | Cause | Fix |
|
||||
|---------|--------|-----|
|
||||
| CUDA version mismatch errors | Wrong CUDA toolkit version | Reinstall CUDA 12.9 using exact installer |
|
||||
| Container registry authentication fails | Invalid or expired GitLab token | Generate new auth token |
|
||||
| SM_121a architecture not recognized | Missing LLVM patches | Verify SM_121a patches applied to LLVM source |
|
||||
| Reduce MAX_JOBS to 1-2, add swap space |
|
||||
| Environment variables not set |
|
||||
|
||||
## Step 4. Cleanup and rollback
|
||||
|
||||
For container approach (non-destructive):
|
||||
|
||||
```bash
|
||||
docker rm $(docker ps -aq --filter ancestor=******:5005/dl/dgx/vllm*)
|
||||
docker rmi ******:5005/dl/dgx/vllm:main-py3.31165712-devel
|
||||
```
|
||||
|
||||
|
||||
To remove CUDA 12.9:
|
||||
|
||||
```bash
|
||||
sudo /usr/local/cuda-12.9/bin/cuda-uninstaller
|
||||
```
|
||||
|
||||
## Step 5. Next steps
|
||||
|
||||
- **Production deployment:** Configure vLLM with your specific model requirements
|
||||
- **Performance tuning:** Adjust batch sizes and memory settings for your workload
|
||||
- **Monitoring:** Set up logging and metrics collection for production use
|
||||
- **Model management:** Explore additional model formats and quantization options
|
||||
|
||||
## Run on two Sparks
|
||||
|
||||
## Step 1. Verify hardware connectivity
|
||||
@ -310,74 +381,3 @@ http://192.168.100.10:8265
|
||||
## - Persistent model caching across restarts
|
||||
## - Alternative quantization methods (FP8, INT4)
|
||||
```
|
||||
|
||||
## Access through terminal
|
||||
|
||||
## Step 1. Pull vLLM container image
|
||||
|
||||
Find the latest container build from https://catalog.ngc.nvidia.com/orgs/nvidia/containers/vllm?version=25.09-py3
|
||||
```
|
||||
docker pull nvcr.io/nvidia/vllm:25.09-py3
|
||||
```
|
||||
|
||||
## Step 2. Test vLLM in container
|
||||
|
||||
Launch the container and start vLLM server with a test model to verify basic functionality.
|
||||
|
||||
```bash
|
||||
docker run -it --gpus all -p 8000:8000 \
|
||||
nvcr.io/nvidia/vllm:25.09-py3 \
|
||||
vllm serve "Qwen/Qwen2.5-Math-1.5B-Instruct"
|
||||
```
|
||||
|
||||
Expected output should include:
|
||||
- Model loading confirmation
|
||||
- Server startup on port 8000
|
||||
- GPU memory allocation details
|
||||
|
||||
In another terminal, test the server:
|
||||
|
||||
```bash
|
||||
curl http://localhost:8000/v1/chat/completions \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"model": "Qwen/Qwen2.5-Math-1.5B-Instruct",
|
||||
"messages": [{"role": "user", "content": "12*17"}],
|
||||
"max_tokens": 500
|
||||
}'
|
||||
```
|
||||
|
||||
Expected response should contain `"content": "204"` or similar mathematical calculation.
|
||||
|
||||
## Step 3. Troubleshooting
|
||||
|
||||
| Symptom | Cause | Fix |
|
||||
|---------|--------|-----|
|
||||
| CUDA version mismatch errors | Wrong CUDA toolkit version | Reinstall CUDA 12.9 using exact installer |
|
||||
| Container registry authentication fails | Invalid or expired GitLab token | Generate new auth token |
|
||||
| SM_121a architecture not recognized | Missing LLVM patches | Verify SM_121a patches applied to LLVM source |
|
||||
| Reduce MAX_JOBS to 1-2, add swap space |
|
||||
| Environment variables not set |
|
||||
|
||||
## Step 4. Cleanup and rollback
|
||||
|
||||
For container approach (non-destructive):
|
||||
|
||||
```bash
|
||||
docker rm $(docker ps -aq --filter ancestor=******:5005/dl/dgx/vllm*)
|
||||
docker rmi ******:5005/dl/dgx/vllm:main-py3.31165712-devel
|
||||
```
|
||||
|
||||
|
||||
To remove CUDA 12.9:
|
||||
|
||||
```bash
|
||||
sudo /usr/local/cuda-12.9/bin/cuda-uninstaller
|
||||
```
|
||||
|
||||
## Step 5. Next steps
|
||||
|
||||
- **Production deployment:** Configure vLLM with your specific model requirements
|
||||
- **Performance tuning:** Adjust batch sizes and memory settings for your workload
|
||||
- **Monitoring:** Set up logging and metrics collection for production use
|
||||
- **Model management:** Explore additional model formats and quantization options
|
||||
|
||||
@ -43,14 +43,14 @@ You will deploy NVIDIA's VSS AI Blueprint on NVIDIA Spark hardware with Blackwel
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- [ ] NVIDIA Spark device with ARM64 architecture and Blackwell GPU
|
||||
- [ ] FastOS 1.81.38 or compatible ARM64 system
|
||||
- [ ] Driver version 580.82.09 installed: `nvidia-smi | grep "Driver Version"`
|
||||
- [ ] CUDA version 13.0 installed: `nvcc --version`
|
||||
- [ ] Docker installed and running: `docker --version && docker compose version`
|
||||
- [ ] Access to NVIDIA Container Registry with NGC API Key
|
||||
- [ ] [Optional] NVIDIA API Key for remote model endpoints (hybrid deployment only)
|
||||
- [ ] Sufficient storage space for video processing (>10GB recommended in `/tmp/`)
|
||||
- NVIDIA Spark device with ARM64 architecture and Blackwell GPU
|
||||
- FastOS 1.81.38 or compatible ARM64 system
|
||||
- Driver version 580.82.09 installed: `nvidia-smi | grep "Driver Version"`
|
||||
- CUDA version 13.0 installed: `nvcc --version`
|
||||
- Docker installed and running: `docker --version && docker compose version`
|
||||
- Access to NVIDIA Container Registry with NGC API Key
|
||||
- [Optional] NVIDIA API Key for remote model endpoints (hybrid deployment only)
|
||||
- Sufficient storage space for video processing (>10GB recommended in `/tmp/`)
|
||||
|
||||
## Ancillary files
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user