Model reference · open weights
tiny-mixtral is an open-weight language model from TitanML. tiny-mixtral (FP32) weighs 494 MB; the smallest configuration that runs it is RTX 3060 12 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
What it is
| Released by | TitanML |
|---|---|
| Released | 2024-04-24 |
| Parameters | 247M |
| VRAM | 494 MB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory | |
|---|---|---|---|---|
| 8K each | 32K each | |||
| RTX 3060 12 GB | 632 | 632 | all 128K | 11.6 GB |
| RTX 4060 Ti 16 GB | 858 | 858 | all 128K | 15.4 GB |
| RTX 3090 24 GB | 1000+ | 1000+ | all 128K | 23.4 GB |
| RTX 4090 24 GB | 1000+ | 1000+ | all 128K | 23.4 GB |
| RTX 5090 32 GB | 1000+ | 1000+ | all 128K | 31.0 GB |
| L40S 48 GB | 1000+ | 1000+ | all 128K | 44.0 GB |
| A100 80 GB | 1000+ | 1000+ | all 128K | 78.2 GB |
| H100 80 GB | 1000+ | 1000+ | all 128K | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | 1000+ | 1000+ | all 128K | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | 1000+ | 1000+ | all 128K | 107 GB |
| H200 141 GB | 1000+ | 1000+ | all 128K | 138 GB |
| B200 180 GB | 1000+ | 1000+ | all 128K | 176 GB |
| Requests at once | 8K tokens each | 32K tokens each |
|---|---|---|
| 1 | 1.0 GB | 1.0 GB |
| 5 | 1.1 GB | 1.1 GB |
| 8 | 1.1 GB | 1.1 GB |
| 16 | 1.3 GB | 1.3 GB |
| 32 | 1.5 GB | 1.5 GB |
| 64 | 2.1 GB | 2.1 GB |
One card, with vLLM's small-card settings.
From the model card
This is a tiny-random mixtral, useful for testing and CI/CD pipelines. It is not trained at all, and not suitable for inferencing in any application.
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.