← CanIRunLocal · All hardware · All models
AutoTrust AI · 27B parameters · chat.
There is no Ollama tag for JEV-27B (AutoTrust, decision), so there is no ollama run command. The fits below use verified Hugging Face weights.
Decision mode (System 1: yes/no, multiple-choice and 0–5 ratings with calibrated probabilities in one forward pass) needs the bundled LoRA adapter. From the model card: hf download autotrust/JEV-27B --local-dir JEV-27B, then vllm serve JEV-27B --enable-lora --max-lora-rank 32 --lora-modules jev-decision=JEV-27B/adapter_vllm --logprobs-mode processed_logprobs --max-model-len 32768 (or the repo's serve_decide.py with the same flags for POST /v1/decide). The plain vllm serve commands below give normal text generation (System 2, the unchanged Qwen3.8-27B). Community GGUF files contain only the Qwen3.8-27B backbone without the decision adapter, so llama.cpp, LM Studio and MLX are not listed.
Yes on 3 machines in this list. The smallest is DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte): BF16, about 55.7 GB, roughly 4 tokens per second, via vLLM. vllm serve autotrust/JEV-27B --max-model-len 8192 --gpu-memory-utilization 0.9
vllm serve autotrust/JEV-27B --max-model-len 8192 --gpu-memory-utilization 0.9vllm serve autotrust/JEV-27B --max-model-len 8192 --gpu-memory-utilization 0.9| Hardware | Runtime | Quant | Memory | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte) | vLLM | Official weights | 55.7 GB | fits in fast memory | ~4 tok/s | vllm serve autotrust/JEV-27B --max-model-len 8192 --gpu-memory-utilization 0.9 |
| 2x DGX Spark 128GB | vLLM | Official weights | 55.7 GB | fits in fast memory | ~5 tok/s | vllm serve autotrust/JEV-27B --tensor-parallel-size 2 --distributed-executor-backend ray --max-model-len 8192 |
| 4x DGX Spark 128GB | vLLM | Official weights | 55.7 GB | fits in fast memory | ~7 tok/s | vllm serve autotrust/JEV-27B --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 2026-10-19.
A project by OnDevice Labs · llms.txt