TinyLlama-1.1B-step-50K-105b
by TinyLlama
1.1B params · text-generation · 134 likes · 14.7k downloads
TinyLlama-1.1B-step-50K-105b is a 1.1B parameter model. At Q4 quantization it requires 1GB of VRAM. It runs comfortably on GeForce RTX 4090 (1193 tok/s), GeForce RTX 5090 (1789 tok/s), M4 Max 128GB (436 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Featherless |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 1193 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 1789 tok/s | comfortable |
| M4 Max 128GB | 128GB | 436 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 218 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 218 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 2186 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 4398 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 1055 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 872 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 847 tok/s | comfortable |