Cloud Deployment

1from langtrain import LoRATrainer
2
3# After training, push to cloud
4trainer = LoRATrainer(model="llama-3.3-8b", output_dir="./model")
5trainer.train("data.jsonl")
6
7# Deploy to Langtrain Cloud
8trainer.push("my-assistant")
9
10# Your model is now available at:
11# POST https://api.langtrain.xyz/v1/chat/completions
12# Model: my-assistant

1import openai
2
3# Use with OpenAI SDK
4client = openai.OpenAI(
5    api_key="your-langtrain-api-key",
6    base_url="https://api.langtrain.xyz/v1"
7)
8
9response = client.chat.completions.create(
10    model="my-assistant",
11    messages=[{"role": "user", "content": "Hello!"}]
12)
13
14print(response.choices[0].message.content)
15
16# Streaming
17stream = client.chat.completions.create(
18    model="my-assistant",
19    messages=[{"role": "user", "content": "Tell me a story"}],
20    stream=True
21)
22for chunk in stream:
23    print(chunk.choices[0].delta.content or "", end="")

1# Export model to Docker
2langtrain export docker --model my-assistant --output ./docker
3
4# Build and run locally
5cd docker
6docker build -t my-assistant:latest .
7docker run -p 8000:8000 --gpus all my-assistant:latest
8
9# Test the endpoint
10curl -X POST http://localhost:8000/v1/chat/completions \
11  -H "Content-Type: application/json" \
12  -d '{"model": "my-assistant", "messages": [{"role": "user", "content": "Hello"}]}'

1# Install Langtrain Helm chart
2helm repo add langtrain https://charts.langtrain.xyz
3helm repo update
4
5# Deploy with GPU support
6helm install my-assistant langtrain/model \
7  --set model.name=my-assistant \
8  --set model.apiKey=$LANGTRAIN_API_KEY \
9  --set resources.limits.nvidia.com/gpu=1 \
10  --set replicas=3
11
12# Expose via ingress
13kubectl apply -f - <<EOF
14apiVersion: networking.k8s.io/v1
15kind: Ingress
16metadata:
17  name: my-assistant
18spec:
19  rules:
20  - host: api.mycompany.com
21    http:
22      paths:
23      - path: /
24        backend:
25          service:
26            name: my-assistant
27            port: 8000
28EOF

1# Langtrain Cloud auto-scaling (via dashboard or API)
2deployment_config = {
3    "model": "my-assistant",
4    "scaling": {
5        "min_replicas": 1,
6        "max_replicas": 10,
7        "target_gpu_utilization": 70,
8        "scale_up_cooldown": "2m",
9        "scale_down_cooldown": "10m"
10    },
11    "instance_type": "gpu-a100-40gb"
12}
13
14# Kubernetes HPA
15kubectl autoscale deployment my-assistant \
16  --min=1 --max=10 \
17  --cpu-percent=70

1# Export to HuggingFace format
2trainer.export("./export", format="huggingface")
3
4# Export to GGUF for llama.cpp
5trainer.export("./export", format="gguf", quantization="q4_k_m")
6
7# Export to ONNX for edge deployment
8trainer.export("./export", format="onnx")
9
10# Upload to HuggingFace Hub
11trainer.push_to_hub("your-username/my-model")