parakeet-tdt-0.6b-v3
by nvidia
627M params · automatic-speech-recognition · 938 likes · 156.2k downloads
parakeet-tdt-0.6b-v3 is a 627M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (2093 tok/s), GeForce RTX 5090 (3140 tok/s), M4 Max 128GB (766 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Together AI |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 2093 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 3140 tok/s | comfortable |
| M4 Max 128GB | 128GB | 766 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 383 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 383 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 3836 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 7716 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 1851 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 1530 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 1486 tok/s | comfortable |