Model reference · open weights
openmoe is an open-weight language model from hpcai-tech. openmoe-8B (BF16) weighs 47.4 GB; the smallest configuration that runs it is 2× RTX 5090 32 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
What it is
| Released by | hpcai-tech |
|---|---|
| Released | 2023-09-05 |
| VRAM | 47.4 GB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory |
|---|---|---|---|
| 2K each | |||
| RTX 3060 12 GB … L40S 48 GB 6 smaller cards | |||
| A100 80 GB | 49 | all 2K | 78.2 GB |
| H100 80 GB | 40 | all 2K | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | 66 | all 2K | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | 89 | all 2K | 107 GB |
| H200 141 GB | 139 | all 2K | 138 GB |
| B200 180 GB | 202 | all 2K | 176 GB |
| 2× RTX 5090 32 GB tensor parallel | 21 | all 2K | 31.0 GB a card |
| 2× L40S 48 GB tensor parallel | 64 | all 2K | 44.0 GB a card |
| 4× RTX 4090 24 GB tensor parallel | 70 | all 2K | 23.4 GB a card |
| 4× RTX 3090 24 GB tensor parallel | 71 | all 2K | 23.4 GB a card |
| Requests at once | 2K tokens each |
|---|---|
| 1 | 48.8 GB |
| 5 | 51.2 GB |
| 8 | 53.0 GB |
| 16 | 57.9 GB |
| 32 | 67.5 GB |
| 64 | 86.9 GB |
One card, with vLLM's small-card settings.