Models

Popular models from Hugging Face that GridPup can serve, grouped by family and size, with the GPU memory each one needs and the cheapest machine that fits it right now. Open a family to compare its sizes and versions (the original, quantized and ONNX versions need less GPU memory) and the GPUs for each.

2,289 families · catalog updated 7 hours ago

• Meta-Llama-3 by NousResearchText generation · 3 models, 5 versions · ↓ 220.7K8B70B
  Smallest needs21.3 GB GPU
  From$0.19/h2x RTX 3060
• OLMoE by allenaiText generation · 4 models · ↓ 215.2K1B
  Smallest needs18.2 GB GPU
  From$0.19/h2x RTX 3060
• Phi-3-mini-128k by microsoftText generation · 2 models · ↓ 211.7K~3.82B
  Smallest needs7.2 GB GPU
  From$0.09/hRTX 2060 SUPER
• MiMo by XiaomiMiMoText generation · 3 models · ↓ 204.5K7B~311B
  Smallest needs21 GB GPU
  From$0.19/h2x RTX 3060
• llama-3 by unslothText generation · 3 models · ↓ 202.3K8B
  Smallest needs9.8 GB GPU

Model details come from Hugging Face and are refreshed daily. GPU memory is GridPup's estimate: the weights, room for an 8,192-token conversation (chat models) and the runtime's own memory. Large models are split across the GPUs of one machine ("4x RTX 4090"). Only machines with a Vast reliability above 95% that download at 400 Mbit/s or more are counted.