Optimized GGUF Models for TinyLlama/TinyLlama-1.1B-Chat-v1.0
This repository contains highly optimized, quantized GGUF models designed for offline-native and edge inference.
📊 Performance Benchmarks (Wikitext-2 Perplexity)
Lower perplexity (PPL) indicates better performance and less degradation.
| Model File | Perplexity Score (PPL) |
|---|
model-fp16.gguf | 8.4441 |
model-q4_k_m.gguf | 8.7433 |
model-q8_0.gguf | 8.4484 |
Auto-generated via the GGUF Conversion & Benchmarking Pipeline.
