Loading…
Loading…
Models / LFM2.5
6 models (230M to 24B) · ↓ 592.2K downloads a month · from $0.08/h on GTX 1660 SUPER right now. Each model's versions do the same job; quantized ones (AWQ, FP8, 4-bit) need less GPU memory and run on cheaper machines.
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|---|---|---|---|
| OriginalLiquidAI/LFM2.5-350M-Base | vLLM | 710 MB | 3.3 GB | $0.10/h RTX 3060 |
| InstructLiquidAI/LFM2.5-350M | vLLM | 710 MB | 3.3 GB | $0.10/h RTX 3060 |
| ONNX (onnx-community)onnx-community/LFM2.5-350M-ONNX | ONNX Runtime | 260 MB | 1.4 GB | $0.08/h GTX 1660 SUPER |
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|---|---|---|---|
| OriginalLiquidAI/LFM2.5-1.2B-Base | vLLM | 2.3 GB | 5.1 GB | $0.10/h RTX 3060 |
| InstructLiquidAI/LFM2.5-1.2B-Instruct | vLLM | 2.3 GB | 5.1 GB | $0.10/h RTX 3060 |
| JP 202606LiquidAI/LFM2.5-1.2B-JP-202606 | vLLM | 2.3 GB | 5.1 GB | $0.10/h RTX 3060 |
| JP 202606 ONNXLiquidAI/LFM2.5-1.2B-JP-202606-ONNX | ONNX Runtime | 740 MB | 2.1 GB | $0.08/h GTX 1660 SUPER |
| ThinkingLiquidAI/LFM2.5-1.2B-Thinking | vLLM | 2.3 GB | 5.1 GB | $0.10/h RTX 3060 |
| JPLiquidAI/LFM2.5-1.2B-JP | vLLM | 2.3 GB | 5.1 GB | $0.10/h RTX 3060 |
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|---|---|---|---|
| OriginalLiquidAI/LFM2.5-2.6B-Base | vLLM | 5.4 GB | 8.8 GB | $0.10/h RTX 3060 |
| InstructLiquidAI/LFM2.5-2.6B | vLLM | 5.4 GB | 8.8 GB | $0.10/h RTX 3060 |
| UNCENSORED ABLITERATED PHILADELPHIA CLASS (KridgeDookie)KridgeDookie/LFM2.5-2.6B-UNCENSORED-ABLITERATED-PHILADELPHIA-CLASS | vLLM | 5.4 GB | 8.8 GB | $0.10/h RTX 3060 |
| Version | Runtime | Weights | GPU memory | Cheapest machine now |
|---|---|---|---|---|
| InstructLiquidAI/LFM2.5-8B-A1B | vLLM | 17 GB | 21.5 GB | $0.18/h 2x RTX 3060 |
Model details come from Hugging Face and are refreshed daily; prices follow the live offers. 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").