refactor: Move Llama 4 Trainium2 deployment to Helm charts

- Remove YAML deployment files from main repo
- Update documentation to use Helm-based deployment
- Add model compilation (tracing) requirement documentation
- Reference ai-on-eks-charts repo for deployment configs

Author: devfloor9

blueprints/inference/llama4-vllm-trn2/README.md

@@ -1,125 +1,31 @@
-# Llama 4 with vLLM on Amazon EKS using Trainium2
+# Llama 4 with vLLM on Trainium2
 
-This blueprint deploys Llama 4 models (Scout and Maverick) using vLLM with NxD Inference on AWS Trainium2 instances.
-
-## Overview
-
-Llama 4 models use a Mixture of Experts (MoE) architecture that requires significant compute resources. This blueprint leverages AWS Trainium2 (trn2) instances with the Neuron SDK for cost-effective inference.
-
-## Model Support
-
-| Model | Parameters | Experts | Instance Required | tensor_parallel_size |
-|-------|------------|---------|-------------------|---------------------|
-| Llama 4 Scout | 17B active / ~109B total | 16 | trn2.48xlarge | 64 |
-| Llama 4 Maverick | 17B active / ~400B total | 128 | trn2.48xlarge | 64 |
-
-## Prerequisites
-
-1. EKS cluster with Trainium2 node support
-2. Neuron device plugin installed
-3. Hugging Face account with Llama 4 model access
-4. Neuron SDK 2.21+ with vLLM-Neuron plugin
+This blueprint has been moved to the [AI on EKS Inference Charts](https://github.com/awslabs/ai-on-eks-charts) repository.
 
 ## Deployment
 
-### Step 1: Export Hugging Face Token
-
-```bash
-export HUGGING_FACE_HUB_TOKEN=$(echo -n "your-hf-token" | base64)
-```
-
-### Step 2: Deploy Llama 4 Scout
-
-```bash
-envsubst < llama4-vllm-trn2-deployment.yaml | kubectl apply -f -
-```
-
-### Step 3: Deploy Open WebUI (Optional)
-
-```bash
-kubectl apply -f open-webui.yaml
-kubectl -n open-webui port-forward svc/open-webui 8080:80
-```
-
-### Step 4: Test with curl
-
-```bash
-kubectl -n llama4-vllm port-forward svc/llama4-vllm-trn2-svc 8000:8000
-
-# Text completion
-curl -X POST http://localhost:8000/v1/chat/completions \
-  -H "Content-Type: application/json" \
-  -d '{
-    "model": "meta-llama/Llama-4-Scout-17B-16E-Instruct",
-    "messages": [{"role": "user", "content": "Hello!"}]
-  }'
-
-# Multimodal (image + text)
-curl -X POST http://localhost:8000/v1/chat/completions \
-  -H "Content-Type: application/json" \
-  -d '{
-    "model": "meta-llama/Llama-4-Scout-17B-16E-Instruct",
-    "messages": [{
-      "role": "user",
-      "content": [
-        {"type": "image_url", "image_url": {"url": "https://httpbin.org/image/png"}},
-        {"type": "text", "text": "Describe this image"}
-      ]
-    }]
-  }'
-```
-
-## Deploying Maverick Model
-
-For the larger Maverick model with 128 experts:
+Deploy using Helm:
 
 ```bash
-envsubst < llama4-vllm-trn2-maverick.yaml | kubectl apply -f -
+# Add the Helm repository
+helm repo add ai-on-eks https://awslabs.github.io/ai-on-eks-charts
+helm repo update
+
+# Deploy Llama 4 Scout on Trainium2
+helm install llama4-scout ai-on-eks/inference-charts \
+  -f https://raw.githubusercontent.com/awslabs/ai-on-eks-charts/main/charts/inference-charts/values-llama-4-scout-17b-vllm-neuron.yaml \
+  --set inference.modelServer.env.NEURON_COMPILED_ARTIFACTS="s3://your-bucket/llama4-neuron-artifacts/"
+
+# Deploy Llama 4 Maverick on Trainium2
+helm install llama4-maverick ai-on-eks/inference-charts \
+  -f https://raw.githubusercontent.com/awslabs/ai-on-eks-charts/main/charts/inference-charts/values-llama-4-maverick-17b-vllm-neuron.yaml \
+  --set inference.modelServer.env.NEURON_COMPILED_ARTIFACTS="s3://your-bucket/llama4-neuron-artifacts/"
 ```
 
-## Key Configuration
-
-### Neuron Configuration
-
-The deployment includes optimized Neuron configuration for Llama 4:
-
-- `tensor_parallel_size=64`: Distributes model across all 64 Neuron cores
-- `context_encoding_buckets`: Optimized bucket sizes for variable input lengths
-- `async_mode=true`: Enables asynchronous execution for better throughput
-- `cp_degree=16`: Context parallelism for efficient attention computation
-
-### Resource Requirements
+## Documentation
 
-| Resource | Scout | Maverick |
-|----------|-------|----------|
-| Neuron Devices | 32 | 32 |
-| CPU | 128-192 | 128-192 |
-| Memory | 512-768 Gi | 512-768 Gi |
-| Model Storage | 1 Ti | 2 Ti |
-
-## Model Compilation (Tracing)
-
-Before deployment, models must be compiled (traced) for Neuron. This is typically done once and the artifacts are stored:
-
-```python
-# Example tracing script (run on trn2 instance)
-import os
-os.environ['NEURON_COMPILED_ARTIFACTS'] = "/models/traced_models/Llama-4-Scout-17B-16E-Instruct"
-
-# Tracing happens automatically on first run with vLLM-Neuron
-```
-
-## Cleanup
-
-```bash
-kubectl delete -f open-webui.yaml
-kubectl delete -f llama4-vllm-trn2-deployment.yaml
-# Or for Maverick:
-kubectl delete -f llama4-vllm-trn2-maverick.yaml
-```
+See the full documentation at: https://awslabs.github.io/ai-on-eks/docs/blueprints/inference/Neuron/llama4-trn2
 
-## References
+## Important Note
 
-- [NxD Inference Llama 4 Tutorial](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/libraries/nxd-inference/tutorials/llama4-tutorial.html)
-- [vLLM-Neuron Plugin](https://github.com/aws-neuron/vllm-neuron)
-- [AWS Neuron SDK](https://awsdocs-neuron.readthedocs-hosted.com/)
+Llama 4 models on Trainium2 require **pre-compiled (traced) model artifacts**. See the [NxD Inference Llama 4 Tutorial](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/libraries/nxd-inference/tutorials/llama4-tutorial.html) for compilation instructions.