GPU Memory Optimization Guide

As a Lead AI Engineer, optimize GPU memory usage for our inference workload.

## Current Setup
{{current_setup}}

## Model Details
- Model: {{model_name}}
- Parameters: {{param_count}}
- Precision: {{precision}}

## GPU Constraints
- GPU type: {{gpu_type}}
- VRAM: {{vram_gb}}GB

## Optimization Goals
{{optimization_goals}}

Analyze optimization strategies:

1. **Quantization**: INT8, INT4, FP16, BF16 trade-offs
2. **Batching**: Optimal batch size for memory/throughput
3. **KV Cache**: Memory estimation and management
4. **Attention**: Flash attention, paged attention options
5. **Tensor Parallelism**: Multi-GPU distribution

Provide:
- Memory calculations for each configuration
- Performance impact estimates
- Implementation recommendations
- Profiling methodology