Model reference · open weights
CASA-Helium1-VL-Shared is an open-weight language model from kyutai. CASA-Helium1-VL-2B-Shared (BF16) weighs 5.4 GB; 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.
Summary of the kyutai/CASA-Helium1-VL-2B-Shared model card, 2026-10-01
What it is
| Released by | kyutai |
|---|---|
| Released | 2026-08-26 |
| Parameters | 2.7B |
| VRAM | 5.4 GB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory |
|---|---|---|---|
| 4K each | |||
| RTX 3060 12 GB | 9 | all 4K | 11.6 GB |
| RTX 4060 Ti 16 GB | 17 | all 4K | 15.4 GB |
| RTX 3090 24 GB | 34 | all 4K | 23.4 GB |
| RTX 4090 24 GB | 34 | all 4K | 23.4 GB |
| RTX 5090 32 GB | 50 | all 4K | 31.0 GB |
| L40S 48 GB | 78 | all 4K | 44.0 GB |
| A100 80 GB | 150 | all 4K | 78.2 GB |
| H100 80 GB | 144 | all 4K | 78.1 GB |
| RTX PRO 6000 Blackwell 96 GB | 177 | all 4K | 93.8 GB |
| DGX Spark (GB10) 128 GB unified | 206 | all 4K | 107 GB |
| H200 141 GB | 271 | all 4K | 138 GB |
| B200 180 GB | 352 | all 4K | 176 GB |
| Requests at once | 4K tokens each |
|---|---|
| 1 | 7.7 GB |
| 5 | 9.6 GB |
| 8 | 11.0 GB |
| 16 | 14.8 GB |
| 32 | 22.3 GB |
| 64 | 37.3 GB |
One card, with vLLM's small-card settings.
From the model card
This repository contains the model weights for CASA-Helium1-VL-2B-Shared, introduced in the paper CASA: Cross-Attention over Self-Attention for Efficient Vision-Language Fusion.
This is a variant of CASA-Helium1-VL-2B where the self-attention and cross-attention layers share the same parameters.
See CASA-Helium1-VL-2B for more information
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.