LLaVA models
9 LLaVA models from llava-hf on Hugging Face, from 894M to 13.4B parameters, published by llava-hf in F16 and BF16, with contexts from 4K to 32K tokens. The smallest official model, llava-onevision-qwen2-0.5b-ov-hf, needs about 2.0 GB of VRAM at its published precision; the cheapest live fit is RTX 3070 at $0.088/hr.
Pick a size
One row per official LLaVA size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.
| Model | Parameters | Native VRAM | FP8 VRAM | INT4 VRAM | Live GPU fit (native) | Est. $/hr | KV cache at 32K |
|---|---|---|---|---|---|---|---|
| llava-onevision-qwen2-0.5b-ov-hf | 894M | 2.0 GB | 1.0 GB | 0.5 GB | RTX 3070 | $0.088/hr | 0.38 GB |
| llava-1.5-7b-hf | 7.1B | 15.8 GB | 7.9 GB | 3.9 GB | V100 | $0.110/hr | not published for this architecture |
| llava-v1.6-mistral-7b-hf | 7.6B | 16.9 GB | 8.5 GB | 4.2 GB | RTX A5000 | $0.176/hr | not published for this architecture |
| llava-1.5-13b-hf | 13.4B | 29.8 GB | 14.9 GB | 7.5 GB | V100 | $0.220/hr | 25.0 GB |
VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.
Official models (6)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| llava-onevision-qwen2-0.5b-ov-hf | 894M | F16 | 2.0 GB | RTX 3070 | $0.088/hr |
| llava-v1.6-vicuna-7b | 7.1B | BF16 | 15.8 GB | RTX 5060 Ti | $0.145/hr |
| llava-1.5-7b-hf | 7.1B | F16 | 15.8 GB | V100 | $0.110/hr |
| llava-v1.6-mistral-7b | 7.6B | BF16 | 16.9 GB | RTX A5000 | $0.176/hr |
| llava-v1.6-mistral-7b-hf | 7.6B | F16 | 16.9 GB | RTX A5000 | $0.176/hr |
| llava-1.5-13b-hf | 13.4B | F16 | 29.8 GB | V100 | $0.220/hr |
Fine-tunes and community models (3)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| llava-med-v1.5-mistral-7b | 7.6B | BF16 | 16.9 GB | RTX A5000 | $0.176/hr |
| llava-onevision-qwen2-7b-ov | 8.0B | BF16 | 17.9 GB | RTX A5000 | $0.176/hr |
| LLaVA-Video-7B-Qwen2 | 8.0B | BF16 | 17.9 GB | RTX A5000 | $0.176/hr |
VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.