← CanIRunLocal · All hardware · All models
IFM · 375B parameters (23B active) · chat, reasoning.
There is no Ollama tag for Jais 3 375B-A23B (MoE), so there is no ollama run command. The fits below use verified Hugging Face weights.
No GGUF or MLX build is published. IFM/Jais-3-375B-A23B-GGUF is not on Hugging Face, and community GGUF searches came up empty, so Ollama, llama.cpp, LM Studio and MLX are not listed. vLLM can load inception42/Jais-3-375B-A23B or IFM/Jais-3-375B-A23B-FP8.
Yes on 1 machine in this list. The smallest is 4x DGX Spark 128GB: FP8, about 392.9 GB, roughly 12 tokens per second, via vLLM. vllm serve IFM/Jais-3-375B-A23B-FP8 --tensor-parallel-size 4 --distributed-executor-backend ray --max-model-len 8192
vllm serve IFM/Jais-3-375B-A23B-FP8 --tensor-parallel-size 4 --distributed-executor-backend ray --max-model-len 8192| Hardware | Runtime | Quant | Memory | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| 4x DGX Spark 128GB | vLLM | FP8 | 392.9 GB | fits in fast memory | ~12 tok/s | vllm serve IFM/Jais-3-375B-A23B-FP8 --tensor-parallel-size 4 --distributed-executor-backend ray --max-model-len 8192 |
Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-20.
A project by OnDevice Labs · llms.txt