@r1z4x/tezgah 1.0.0

package/skills/ai-research/12-inference-serving/vllm/references/quantization.md

@@ -0,0 +1,285 @@
+<!-- vendored from orchestra-research/AI-research-SKILLs@773a52944ba4747a18bd4ae9ade53fff041adcbc 12-inference-serving/vllm/references/quantization.md; MIT (c) Orchestra Research -->
+# Quantization Guide
+
+## Contents
+- Quantization methods comparison
+- AWQ setup and usage
+- GPTQ setup and usage
+- FP8 quantization (H100)
+- Model preparation
+- Accuracy vs compression trade-offs
+
+## Quantization methods comparison
+
+| Method | Compression | Accuracy Loss | Speed | Best For |
+|--------|-------------|---------------|-------|----------|
+| **AWQ** | 4-bit (75%) | <1% | Fast | 70B models, production |
+| **GPTQ** | 4-bit (75%) | 1-2% | Fast | Wide model support |
+| **FP8** | 8-bit (50%) | <0.5% | Fastest | H100 GPUs only |
+| **SqueezeLLM** | 3-4 bit (75-80%) | 2-3% | Medium | Extreme compression |
+
+**Recommendation**:
+- **Production**: Use AWQ for 70B models
+- **H100 GPUs**: Use FP8 for best speed
+- **Maximum compatibility**: Use GPTQ
+- **Extreme compression**: Use SqueezeLLM
+
+## AWQ setup and usage
+
+**AWQ** (Activation-aware Weight Quantization) achieves best accuracy at 4-bit.
+
+**Step 1: Find pre-quantized model**
+
+Search HuggingFace for AWQ models:
+```bash
+# Example: TheBloke/Llama-2-70B-AWQ
+# Example: TheBloke/Mixtral-8x7B-Instruct-v0.1-AWQ
+```
+
+**Step 2: Launch with AWQ**
+
+```bash
+vllm serve TheBloke/Llama-2-70B-AWQ \
+  --quantization awq \
+  --tensor-parallel-size 1 \
+  --gpu-memory-utilization 0.95
+```
+
+**Memory savings**:
+```
+Llama 2 70B fp16: 140GB VRAM (4x A100 needed)
+Llama 2 70B AWQ: 35GB VRAM (1x A100 40GB)
+= 4x memory reduction
+```
+
+**Step 3: Verify performance**
+
+Test that outputs are acceptable:
+```python
+from openai import OpenAI
+
+client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
+
+# Test complex reasoning
+response = client.chat.completions.create(
+    model="TheBloke/Llama-2-70B-AWQ",
+    messages=[{"role": "user", "content": "Explain quantum entanglement"}]
+)
+
+print(response.choices[0].message.content)
+# Verify quality matches your requirements
+```
+
+**Quantize your own model** (requires GPU with 80GB+ VRAM):
+
+```python
+from awq import AutoAWQForCausalLM
+from transformers import AutoTokenizer
+
+model_path = "meta-llama/Llama-2-70b-hf"
+quant_path = "llama-2-70b-awq"
+
+# Load model
+model = AutoAWQForCausalLM.from_pretrained(model_path)
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+
+# Quantize
+quant_config = {"zero_point": True, "q_group_size": 128, "w_bit": 4}
+model.quantize(tokenizer, quant_config=quant_config)
+
+# Save
+model.save_quantized(quant_path)
+tokenizer.save_pretrained(quant_path)
+```
+
+## GPTQ setup and usage
+
+**GPTQ** has widest model support and good compression.
+
+**Step 1: Find GPTQ model**
+
+```bash
+# Example: TheBloke/Llama-2-13B-GPTQ
+# Example: TheBloke/CodeLlama-34B-GPTQ
+```
+
+**Step 2: Launch with GPTQ**
+
+```bash
+vllm serve TheBloke/Llama-2-13B-GPTQ \
+  --quantization gptq \
+  --dtype float16
+```
+
+**GPTQ configuration options**:
+```bash
+# Specify GPTQ parameters if needed
+vllm serve MODEL \
+  --quantization gptq \
+  --gptq-act-order \  # Activation ordering
+  --dtype float16
+```
+
+**Quantize your own model**:
+
+```python
+from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
+from transformers import AutoTokenizer
+
+model_name = "meta-llama/Llama-2-13b-hf"
+quantized_name = "llama-2-13b-gptq"
+
+# Load model
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoGPTQForCausalLM.from_pretrained(model_name, quantize_config)
+
+# Prepare calibration data
+calib_data = [...]  # List of sample texts
+
+# Quantize
+quantize_config = BaseQuantizeConfig(
+    bits=4,
+    group_size=128,
+    desc_act=True
+)
+model.quantize(calib_data)
+
+# Save
+model.save_quantized(quantized_name)
+```
+
+## FP8 quantization (H100)
+
+**FP8** (8-bit floating point) offers best speed on H100 GPUs with minimal accuracy loss.
+
+**Requirements**:
+- H100 or H800 GPU
+- CUDA 12.3+ (12.8 recommended)
+- Hopper architecture support
+
+**Step 1: Enable FP8**
+
+```bash
+vllm serve meta-llama/Llama-3-70B-Instruct \
+  --quantization fp8 \
+  --tensor-parallel-size 2
+```
+
+**Performance gains on H100**:
+```
+fp16: 180 tokens/sec
+FP8: 320 tokens/sec
+= 1.8x speedup
+```
+
+**Step 2: Verify accuracy**
+
+FP8 typically has <0.5% accuracy degradation:
+```python
+# Run evaluation suite
+# Compare FP8 vs FP16 on your tasks
+# Verify acceptable accuracy
+```
+
+**Dynamic FP8 quantization** (no pre-quantized model needed):
+
+```bash
+# vLLM automatically quantizes at runtime
+vllm serve MODEL --quantization fp8
+# No model preparation required
+```
+
+## Model preparation
+
+**Pre-quantized models (easiest)**:
+
+1. Search HuggingFace: `[model name] AWQ` or `[model name] GPTQ`
+2. Download or use directly: `TheBloke/[Model]-AWQ`
+3. Launch with appropriate `--quantization` flag
+
+**Quantize your own model**:
+
+**AWQ**:
+```bash
+# Install AutoAWQ
+pip install autoawq
+
+# Run quantization script
+python quantize_awq.py --model MODEL --output OUTPUT
+```
+
+**GPTQ**:
+```bash
+# Install AutoGPTQ
+pip install auto-gptq
+
+# Run quantization script
+python quantize_gptq.py --model MODEL --output OUTPUT
+```
+
+**Calibration data**:
+- Use 128-512 diverse examples from target domain
+- Representative of production inputs
+- Higher quality calibration = better accuracy
+
+## Accuracy vs compression trade-offs
+
+**Empirical results** (Llama 2 70B on MMLU benchmark):
+
+| Quantization | Accuracy | Memory | Speed | Production-Ready |
+|--------------|----------|--------|-------|------------------|
+| FP16 (baseline) | 100% | 140GB | 1.0x | ✅ (if memory available) |
+| FP8 | 99.5% | 70GB | 1.8x | ✅ (H100 only) |
+| AWQ 4-bit | 99.0% | 35GB | 1.5x | ✅ (best for 70B) |
+| GPTQ 4-bit | 98.5% | 35GB | 1.5x | ✅ (good compatibility) |
+| SqueezeLLM 3-bit | 96.0% | 26GB | 1.3x | ⚠️ (check accuracy) |
+
+**When to use each**:
+
+**No quantization (FP16)**:
+- Have sufficient GPU memory
+- Need absolute best accuracy
+- Model <13B parameters
+
+**FP8**:
+- Using H100/H800 GPUs
+- Need best speed with minimal accuracy loss
+- Production deployment
+
+**AWQ 4-bit**:
+- Need to fit 70B model in 40GB GPU
+- Production deployment
+- <1% accuracy loss acceptable
+
+**GPTQ 4-bit**:
+- Wide model support needed
+- Not on H100 (use FP8 instead)
+- 1-2% accuracy loss acceptable
+
+**Testing strategy**:
+
+1. **Baseline**: Measure FP16 accuracy on your evaluation set
+2. **Quantize**: Create quantized version
+3. **Evaluate**: Compare quantized vs baseline on same tasks
+4. **Decide**: Accept if degradation < threshold (typically 1-2%)
+
+**Example evaluation**:
+```python
+from evaluate import load_evaluation_suite
+
+# Run on FP16 baseline
+baseline_score = evaluate(model_fp16, eval_suite)
+
+# Run on quantized
+quant_score = evaluate(model_awq, eval_suite)
+
+# Compare
+degradation = (baseline_score - quant_score) / baseline_score * 100
+print(f"Accuracy degradation: {degradation:.2f}%")
+
+# Decision
+if degradation < 1.0:
+    print("✅ Quantization acceptable for production")
+else:
+    print("⚠️ Review accuracy loss")
+```

package/skills/ai-research/12-inference-serving/vllm/references/server-deployment.md

@@ -0,0 +1,256 @@
+<!-- vendored from orchestra-research/AI-research-SKILLs@773a52944ba4747a18bd4ae9ade53fff041adcbc 12-inference-serving/vllm/references/server-deployment.md; MIT (c) Orchestra Research -->
+# Server Deployment Patterns
+
+## Contents
+- Docker deployment
+- Kubernetes deployment
+- Load balancing with Nginx
+- Multi-node distributed serving
+- Production configuration examples
+- Health checks and monitoring
+
+## Docker deployment
+
+**Basic Dockerfile**:
+```dockerfile
+FROM nvidia/cuda:12.1.0-devel-ubuntu22.04
+
+RUN apt-get update && apt-get install -y python3-pip
+RUN pip install vllm
+
+EXPOSE 8000
+
+CMD ["vllm", "serve", "meta-llama/Llama-3-8B-Instruct", \
+     "--host", "0.0.0.0", "--port", "8000", \
+     "--gpu-memory-utilization", "0.9"]
+```
+
+**Build and run**:
+```bash
+docker build -t vllm-server .
+docker run --gpus all -p 8000:8000 vllm-server
+```
+
+**Docker Compose** (with metrics):
+```yaml
+version: '3.8'
+services:
+  vllm:
+    image: vllm/vllm-openai:latest
+    command: >
+      --model meta-llama/Llama-3-8B-Instruct
+      --gpu-memory-utilization 0.9
+      --enable-metrics
+      --metrics-port 9090
+    ports:
+      - "8000:8000"
+      - "9090:9090"
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities: [gpu]
+```
+
+## Kubernetes deployment
+
+**Deployment manifest**:
+```yaml
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: vllm-server
+spec:
+  replicas: 2
+  selector:
+    matchLabels:
+      app: vllm
+  template:
+    metadata:
+      labels:
+        app: vllm
+    spec:
+      containers:
+      - name: vllm
+        image: vllm/vllm-openai:latest
+        args:
+          - "--model=meta-llama/Llama-3-8B-Instruct"
+          - "--gpu-memory-utilization=0.9"
+          - "--enable-prefix-caching"
+        resources:
+          limits:
+            nvidia.com/gpu: 1
+        ports:
+        - containerPort: 8000
+          name: http
+        - containerPort: 9090
+          name: metrics
+        readinessProbe:
+          httpGet:
+            path: /health
+            port: 8000
+          initialDelaySeconds: 30
+          periodSeconds: 10
+        livenessProbe:
+          httpGet:
+            path: /health
+            port: 8000
+          initialDelaySeconds: 60
+          periodSeconds: 30
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: vllm-service
+spec:
+  selector:
+    app: vllm
+  ports:
+  - port: 8000
+    targetPort: 8000
+    name: http
+  - port: 9090
+    targetPort: 9090
+    name: metrics
+  type: LoadBalancer
+```
+
+## Load balancing with Nginx
+
+**Nginx configuration**:
+```nginx
+upstream vllm_backend {
+    least_conn;  # Route to least-loaded server
+    server localhost:8001;
+    server localhost:8002;
+    server localhost:8003;
+}
+
+server {
+    listen 80;
+
+    location / {
+        proxy_pass http://vllm_backend;
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+
+        # Timeouts for long-running inference
+        proxy_read_timeout 300s;
+        proxy_connect_timeout 75s;
+    }
+
+    # Metrics endpoint
+    location /metrics {
+        proxy_pass http://localhost:9090/metrics;
+    }
+}
+```
+
+**Start multiple vLLM instances**:
+```bash
+# Terminal 1
+vllm serve MODEL --port 8001 --tensor-parallel-size 1
+
+# Terminal 2
+vllm serve MODEL --port 8002 --tensor-parallel-size 1
+
+# Terminal 3
+vllm serve MODEL --port 8003 --tensor-parallel-size 1
+
+# Start Nginx
+nginx -c /path/to/nginx.conf
+```
+
+## Multi-node distributed serving
+
+For models too large for single node:
+
+**Node 1** (master):
+```bash
+export MASTER_ADDR=192.168.1.10
+export MASTER_PORT=29500
+export RANK=0
+export WORLD_SIZE=2
+
+vllm serve meta-llama/Llama-2-70b-hf \
+  --tensor-parallel-size 8 \
+  --pipeline-parallel-size 2
+```
+
+**Node 2** (worker):
+```bash
+export MASTER_ADDR=192.168.1.10
+export MASTER_PORT=29500
+export RANK=1
+export WORLD_SIZE=2
+
+vllm serve meta-llama/Llama-2-70b-hf \
+  --tensor-parallel-size 8 \
+  --pipeline-parallel-size 2
+```
+
+## Production configuration examples
+
+**High throughput** (batch-heavy workload):
+```bash
+vllm serve MODEL \
+  --max-num-seqs 512 \
+  --gpu-memory-utilization 0.95 \
+  --enable-prefix-caching \
+  --trust-remote-code
+```
+
+**Low latency** (interactive workload):
+```bash
+vllm serve MODEL \
+  --max-num-seqs 64 \
+  --gpu-memory-utilization 0.85 \
+  --enable-chunked-prefill
+```
+
+**Memory-constrained** (40GB GPU for 70B model):
+```bash
+vllm serve TheBloke/Llama-2-70B-AWQ \
+  --quantization awq \
+  --tensor-parallel-size 1 \
+  --gpu-memory-utilization 0.95 \
+  --max-model-len 4096
+```
+
+## Health checks and monitoring
+
+**Health check endpoint**:
+```bash
+curl http://localhost:8000/health
+# Returns: {"status": "ok"}
+```
+
+**Readiness check** (wait for model loaded):
+```bash
+#!/bin/bash
+until curl -f http://localhost:8000/health; do
+    echo "Waiting for vLLM to be ready..."
+    sleep 5
+done
+echo "vLLM is ready!"
+```
+
+**Prometheus scraping**:
+```yaml
+# prometheus.yml
+scrape_configs:
+  - job_name: 'vllm'
+    static_configs:
+      - targets: ['localhost:9090']
+    metrics_path: '/metrics'
+    scrape_interval: 15s
+```
+
+**Grafana dashboard** (key metrics):
+- Requests per second: `rate(vllm_request_success_total[5m])`
+- TTFT p50: `histogram_quantile(0.5, vllm_time_to_first_token_seconds_bucket)`
+- TTFT p99: `histogram_quantile(0.99, vllm_time_to_first_token_seconds_bucket)`
+- GPU cache usage: `vllm_gpu_cache_usage_perc`
+- Active requests: `vllm_num_requests_running`