Serving Llms Vllm

vLLM - High-Performance LLM Serving

Quick start

vLLM achieves 24x higher throughput than standard transformers through PagedAttention (block-based KV cache) and continuous batching (mixing prefill/decode requests).

Installation:

bashCopy code
pip install vllm

Basic offline inference:

pythonCopy code
from vllm import LLM, SamplingParams

llm = LLM(model="meta-llama/Llama-3-8B-Instruct")
sampling = SamplingParams(temperature=0.7, max_tokens=256)

outputs = llm.generate(["Explain quantum computing"], sampling)
print(outputs[0].outputs[0].text)

OpenAI-compatible server:

bashCopy code
vllm serve meta-llama/Llama-3-8B-Instruct

# Query with OpenAI SDK
python -c "
from openai import OpenAI
client = OpenAI(base_url='http://localhost:8000/v1', api_key='EMPTY')
print(client.chat.completions.create(
    model='meta-llama/Llama-3-8B-Instruct',
    messages=[{'role': 'user', 'content': 'Hello!'}]
).choices[0].message.content)
"

Common workflows

Workflow 1: Production API deployment

Copy this checklist and track progress:

Copy code
Deployment Progress:
- [ ] Step 1: Configure server settings
- [ ] Step 2: Test with limited traffic
- [ ] Step 3: Enable monitoring
- [ ] Step 4: Deploy to production
- [ ] Step 5: Verify performance metrics

Step 1: Configure server settings

Choose configuration based on your model size:

bashCopy code
# For 7B-13B models on single GPU
vllm serve meta-llama/Llama-3-8B-Instruct \
  --gpu-memory-utilization 0.9 \
  --max-model-len 8192 \
  --port 8000

# For 30B-70B models with tensor parallelism
vllm serve meta-llama/Llama-2-70b-hf \
  --tensor-parallel-size 4 \
  --gpu-memory-utilization 0.9 \
  --quantization awq \
  --port 8000

# For production with caching and metrics
vllm serve meta-llama/Llama-3-8B-Instruct \
  --gpu-memory-utilization 0.9 \
  --enable-prefix-caching \
  --enable-metrics \
  --metrics-port 9090 \
  --port 8000 \
  --host 0.0.0.0

Step 2: Test with limited traffic

Run load test before production:

bashCopy code
# Install load testing tool
pip install locust

# Create test_load.py with sample requests
# Run: locust -f test_load.py --host http://localhost:8000

Verify TTFT (time to first token) < 500ms and throughput > 100 req/sec.

Step 3: Enable monitoring

vLLM exposes Prometheus metrics on port 9090:

bashCopy code
curl http://localhost:9090/metrics | grep vllm

Key metrics to monitor:

• vllm:time_to_first_token_seconds - Latency
• vllm:num_requests_running - Active requests
• vllm:gpu_cache_usage_perc - KV cache utilization

Step 4: Deploy to production

Use Docker for consistent deployment:

bashCopy code
# Run vLLM in Docker
docker run --gpus all -p 8000:8000 \
  vllm/vllm-openai:latest \
  --model meta-llama/Llama-3-8B-Instruct \
  --gpu-memory-utilization 0.9 \
  --enable-prefix-caching

Step 5: Verify performance metrics

Check that deployment meets targets:

• TTFT < 500ms (for short prompts)
• Throughput > target req/sec
• GPU utilization > 80%
• No OOM errors in logs

Workflow 2: Offline batch inference

For processing large datasets without server overhead.

Copy this checklist:

Copy code
Batch Processing:
- [ ] Step 1: Prepare input data
- [ ] Step 2: Configure LLM engine
- [ ] Step 3: Run batch inference
- [ ] Step 4: Process results

Step 1: Prepare input data

pythonCopy code
# Load prompts from file
prompts = []
with open("prompts.txt") as f:
    prompts = [line.strip() for line in f]

print(f"Loaded {len(prompts)} prompts")

Step 2: Configure LLM engine

pythonCopy code
from vllm import LLM, SamplingParams

llm = LLM(
    model="meta-llama/Llama-3-8B-Instruct",
    tensor_parallel_size=2,  # Use 2 GPUs
    gpu_memory_utilization=0.9,
    max_model_len=4096
)

sampling = SamplingParams(
    temperature=0.7,
    top_p=0.95,
    max_tokens=512,
    stop=["</s>", "\n\n"]
)

Step 3: Run batch inference

vLLM automatically batches requests for efficiency:

pythonCopy code
# Process all prompts in one call
outputs = llm.generate(prompts, sampling)

# vLLM handles batching internally
# No need to manually chunk prompts

Step 4: Process results

pythonCopy code
# Extract generated text
results = []
for output in outputs:
    prompt = output.prompt
    generated = output.outputs[0].text
    results.append({
        "prompt": prompt,
        "generated": generated,
        "tokens": len(output.outputs[0].token_ids)
    })

# Save to file
import json
with open("results.jsonl", "w") as f:
    for result in results:
        f.write(json.dumps(result) + "\n")

print(f"Processed {len(results)} prompts")

Workflow 3: Quantized model serving

Fit large models in limited GPU memory.

Copy code
Quantization Setup:
- [ ] Step 1: Choose quantization method
- [ ] Step 2: Find or create quantized model
- [ ] Step 3: Launch with quantization flag
- [ ] Step 4: Verify accuracy

Step 1: Choose quantization method

• AWQ: Best for 70B models, minimal accuracy loss
• GPTQ: Wide model support, good compression
• FP8: Fastest on H100 GPUs

Step 2: Find or create quantized model

Use pre-quantized models from HuggingFace:

bashCopy code
# Search for AWQ models
# Example: TheBloke/Llama-2-70B-AWQ

Step 3: Launch with quantization flag

bashCopy code
# Using pre-quantized model
vllm serve TheBloke/Llama-2-70B-AWQ \
  --quantization awq \
  --tensor-parallel-size 1 \
  --gpu-memory-utilization 0.95

# Results: 70B model in ~40GB VRAM

Step 4: Verify accuracy

Test outputs match expected quality:

pythonCopy code
# Compare quantized vs non-quantized responses
# Verify task-specific performance unchanged

When to use vs alternatives

Use vLLM when:

• Deploying production LLM APIs (100+ req/sec)
• Serving OpenAI-compatible endpoints
• Limited GPU memory but need large models
• Multi-user applications (chatbots, assistants)
• Need low latency with high throughput

Use alternatives instead:

• llama.cpp: CPU/edge inference, single-user
• HuggingFace transformers: Research, prototyping, one-off generation
• TensorRT-LLM: NVIDIA-only, need absolute maximum performance
• Text-Generation-Inference: Already in HuggingFace ecosystem

Common issues

Issue: Out of memory during model loading

Reduce memory usage:

bashCopy code
vllm serve MODEL \
  --gpu-memory-utilization 0.7 \
  --max-model-len 4096

Or use quantization:

bashCopy code
vllm serve MODEL --quantization awq

Issue: Slow first token (TTFT > 1 second)

Enable prefix caching for repeated prompts:

bashCopy code
vllm serve MODEL --enable-prefix-caching

For long prompts, enable chunked prefill:

bashCopy code
vllm serve MODEL --enable-chunked-prefill

Issue: Model not found error

Use --trust-remote-code for custom models:

bashCopy code
vllm serve MODEL --trust-remote-code

Issue: Low throughput (<50 req/sec)

Increase concurrent sequences:

bashCopy code
vllm serve MODEL --max-num-seqs 512

Check GPU utilization with nvidia-smi - should be >80%.

Issue: Inference slower than expected

Verify tensor parallelism uses power of 2 GPUs:

bashCopy code
vllm serve MODEL --tensor-parallel-size 4  # Not 3

Enable speculative decoding for faster generation:

bashCopy code
vllm serve MODEL --speculative-model DRAFT_MODEL

Advanced topics

Server deployment patterns: See references/server-deployment.md for Docker, Kubernetes, and load balancing configurations.

Performance optimization: See references/optimization.md for PagedAttention tuning, continuous batching details, and benchmark results.

Quantization guide: See references/quantization.md for AWQ/GPTQ/FP8 setup, model preparation, and accuracy comparisons.

Troubleshooting: See references/troubleshooting.md for detailed error messages, debugging steps, and performance diagnostics.

Hardware requirements

• Small models (7B-13B): 1x A10 (24GB) or A100 (40GB)
• Medium models (30B-40B): 2x A100 (40GB) with tensor parallelism
• Large models (70B+): 4x A100 (40GB) or 2x A100 (80GB), use AWQ/GPTQ

Supported platforms: NVIDIA (primary), AMD ROCm, Intel GPUs, TPUs

Resources

• Official docs: https://docs.vllm.ai
• GitHub: https://github.com/vllm-project/vllm
• Paper: "Efficient Memory Management for Large Language Model Serving with PagedAttention" (SOSP 2023)
• Community: https://discuss.vllm.ai