Update README.md

README.md

@@ -1,3 +1,24 @@
 ---
 license: mit
+tags:
+- kv-cache
+- quantization
+- efficient-inference
+- llm-optimization
+- memory-optimization
+- inference
 ---
+
+# KV-Cache-Compression-Lab: Systematic Analysis of KV-Cache Quantization for Efficient LLM Inference
+
+**Author:** Samudhan S (@samudhan2003)  
+**Portfolio:** Part of 5-project Efficient Inference Series
+
+## Abstract
+
+The Key-Value (KV) cache is a major memory bottleneck in autoregressive LLM inference, scaling linearly with sequence length and batch size. This lab presents a systematic empirical analysis of KV-cache compression techniques (FP16 baseline vs INT8 vs INT4) and their interaction with weight quantization, measuring the trade-off between VRAM reduction, latency improvement, and quality degradation on sub-1B scale models.
+
+## Motivation
+
+During decoding, KV-cache memory often exceeds model weight memory:
+- For Llama-2