Athene-V2-Chat
by Nexusflow
72.7B params · text-generation · 294 likes · 883 downloads
Athene-V2-Chat is a 72.7B parameter model. At Q4 quantization it requires 36GB of VRAM. It runs comfortably on A100 PCIe 80 GB (33 tok/s), H100 SXM5 80 GB (66 tok/s), H200 SXM 141 GB (96 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Featherless |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| M4 Max 128GB | 128GB | 6 tok/s | tight |
| M4 Pro 48GB | 48GB | 3 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 33 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 66 tok/s | comfortable |
| M4 Max 64GB | 64GB | 6 tok/s | tight |
| H200 SXM 141 GB | 141GB | 96 tok/s | comfortable |
| M2 Ultra 192GB | 192GB | 9 tok/s | tight |