← CanIRunLocal · All hardware · All models

Can I run Laya 421M (Convai, decision) locally?

Convai Innovations · 0.421B parameters · chat.

Ollama weight files: BF16 0.84 GB (laya:421m-english-bf16).

Yes on 85 machines in this list. The smallest is M1 8GB: BF16, about 1.5 GB, roughly 65 tokens per second, via Ollama. ollama run laya:421m-english-bf16

DGX Spark 128GB: BF16, ~1.5 GB, fits in fast memory, ~244 tok/s — ollama run laya:421m-english-bf16
Mac Studio M5 Max 128GB: BF16, ~1.5 GB, fits in fast memory, ~585 tok/s — ollama run laya:421m-english-bf16
Mac Studio M5 Ultra 256GB: BF16, ~1.5 GB, fits in fast memory, ~1143 tok/s — ollama run laya:421m-english-bf16

Setup on DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte)

Ollama: ollama run laya:421m-english-bf16
LM Studio: lms get ggml-org/Laya-GGUF@Q8_0
llama.cpp: llama-server -hf ggml-org/Laya-GGUF:Q8_0 -c 8192
vLLM: vllm serve convaiinnovations/laya --max-model-len 8192 --gpu-memory-utilization 0.9

Open Laya 421M (Convai, decision) in the checker

Hardware that fits

Hardware	Runtime	Quant	Memory	Fit	Speed	Command
M1 8GB	Ollama	BF16	1.5 GB	fits in fast memory	~65 tok/s	`ollama run laya:421m-english-bf16`
M2 8GB	Ollama	BF16	1.5 GB	fits in fast memory	~95 tok/s	`ollama run laya:421m-english-bf16`
M3 8GB	Ollama	BF16	1.5 GB	fits in fast memory	~95 tok/s	`ollama run laya:421m-english-bf16`
RTX 3060 8GB	Ollama	BF16	1.5 GB	fits in fast memory	~171 tok/s	`ollama run laya:421m-english-bf16`
RTX 3060 Ti 8GB	Ollama	BF16	1.5 GB	fits in fast memory	~320 tok/s	`ollama run laya:421m-english-bf16`
RTX 3070 8GB	Ollama	BF16	1.5 GB	fits in fast memory	~320 tok/s	`ollama run laya:421m-english-bf16`
RTX 4060 8GB	Ollama	BF16	1.5 GB	fits in fast memory	~194 tok/s	`ollama run laya:421m-english-bf16`
RTX 4060 Ti 8GB	Ollama	BF16	1.5 GB	fits in fast memory	~206 tok/s	`ollama run laya:421m-english-bf16`
RX 7600 8GB	Ollama	BF16	1.5 GB	fits in fast memory	~206 tok/s	`ollama run laya:421m-english-bf16`
RTX 3080 10GB	Ollama	BF16	1.5 GB	fits in fast memory	~543 tok/s	`ollama run laya:421m-english-bf16`
M1 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~65 tok/s	`ollama run laya:421m-english-bf16`
M1 Pro 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~190 tok/s	`ollama run laya:421m-english-bf16`
M2 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~95 tok/s	`ollama run laya:421m-english-bf16`
M2 Pro 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~190 tok/s	`ollama run laya:421m-english-bf16`
M3 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~95 tok/s	`ollama run laya:421m-english-bf16`
M4 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~114 tok/s	`ollama run laya:421m-english-bf16`
RTX 3060 12GB	Ollama	BF16	1.5 GB	fits in fast memory	~257 tok/s	`ollama run laya:421m-english-bf16`
RTX 3080 12GB	Ollama	BF16	1.5 GB	fits in fast memory	~651 tok/s	`ollama run laya:421m-english-bf16`
RTX 4070 12GB	Ollama	BF16	1.5 GB	fits in fast memory	~360 tok/s	`ollama run laya:421m-english-bf16`
RTX 5070 12GB	Ollama	BF16	1.5 GB	fits in fast memory	~480 tok/s	`ollama run laya:421m-english-bf16`
M3 Pro 18GB	Ollama	BF16	1.5 GB	fits in fast memory	~143 tok/s	`ollama run laya:421m-english-bf16`
RTX 4060 Ti 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~206 tok/s	`ollama run laya:421m-english-bf16`
RTX 4070 Ti Super 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~480 tok/s	`ollama run laya:421m-english-bf16`
RTX 4080 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~512 tok/s	`ollama run laya:421m-english-bf16`
RTX 5080 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~686 tok/s	`ollama run laya:421m-english-bf16`
RX 7800 XT 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~446 tok/s	`ollama run laya:421m-english-bf16`
RX 9070 XT 16GB	Ollama	BF16	1.5 GB	fits in fast memory	~457 tok/s	`ollama run laya:421m-english-bf16`
M2 24GB	Ollama	BF16	1.5 GB	fits in fast memory	~95 tok/s	`ollama run laya:421m-english-bf16`
M3 24GB	Ollama	BF16	1.5 GB	fits in fast memory	~95 tok/s	`ollama run laya:421m-english-bf16`
M4 24GB	Ollama	BF16	1.5 GB	fits in fast memory	~114 tok/s	`ollama run laya:421m-english-bf16`
M4 Pro 24GB	Ollama	BF16	1.5 GB	fits in fast memory	~260 tok/s	`ollama run laya:421m-english-bf16`
RTX 3090 24GB	Ollama	BF16	1.5 GB	fits in fast memory	~669 tok/s	`ollama run laya:421m-english-bf16`
RTX 4090 24GB	Ollama	BF16	1.5 GB	fits in fast memory	~720 tok/s	`ollama run laya:421m-english-bf16`
RX 7900 XTX 24GB	Ollama	BF16	1.5 GB	fits in fast memory	~686 tok/s	`ollama run laya:421m-english-bf16`
M1 Max (24c GPU) 32GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M1 Max (32c GPU) 32GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M1 Pro 32GB	Ollama	BF16	1.5 GB	fits in fast memory	~190 tok/s	`ollama run laya:421m-english-bf16`
M2 Max (30c GPU) 32GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M2 Max (38c GPU) 32GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M2 Pro 32GB	Ollama	BF16	1.5 GB	fits in fast memory	~190 tok/s	`ollama run laya:421m-english-bf16`
M4 32GB	Ollama	BF16	1.5 GB	fits in fast memory	~114 tok/s	`ollama run laya:421m-english-bf16`
M3 Max (30c GPU) 36GB	Ollama	BF16	1.5 GB	fits in fast memory	~286 tok/s	`ollama run laya:421m-english-bf16`
M3 Pro 36GB	Ollama	BF16	1.5 GB	fits in fast memory	~143 tok/s	`ollama run laya:421m-english-bf16`
M4 Max (32c GPU) 36GB	Ollama	BF16	1.5 GB	fits in fast memory	~390 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M4 Max (32c GPU) 36GB	Ollama	BF16	1.5 GB	fits in fast memory	~390 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M5 Max (32c GPU) 36GB	Ollama	BF16	1.5 GB	fits in fast memory	~438 tok/s	`ollama run laya:421m-english-bf16`
RTX 5090 32GB	Ollama	BF16	1.5 GB	fits in fast memory	~1280 tok/s	`ollama run laya:421m-english-bf16`
M3 Max (40c GPU) 48GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M4 Max (40c GPU) 48GB	Ollama	BF16	1.5 GB	fits in fast memory	~520 tok/s	`ollama run laya:421m-english-bf16`
M4 Pro 48GB	Ollama	BF16	1.5 GB	fits in fast memory	~260 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M4 Max (40c GPU) 48GB	Ollama	BF16	1.5 GB	fits in fast memory	~520 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M5 Max (40c GPU) 48GB	Ollama	BF16	1.5 GB	fits in fast memory	~585 tok/s	`ollama run laya:421m-english-bf16`
M1 Max (32c GPU) 64GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M2 Max (30c GPU) 64GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M2 Max (38c GPU) 64GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M2 Ultra 64GB	Ollama	BF16	1.5 GB	fits in fast memory	~762 tok/s	`ollama run laya:421m-english-bf16`
M3 Max (40c GPU) 64GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M4 Max (40c GPU) 64GB	Ollama	BF16	1.5 GB	fits in fast memory	~520 tok/s	`ollama run laya:421m-english-bf16`
M4 Pro 64GB	Ollama	BF16	1.5 GB	fits in fast memory	~260 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M4 Max (40c GPU) 64GB	Ollama	BF16	1.5 GB	fits in fast memory	~520 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M5 Max (40c GPU) 64GB	Ollama	BF16	1.5 GB	fits in fast memory	~585 tok/s	`ollama run laya:421m-english-bf16`
M2 Max (38c GPU) 96GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M3 Max (30c GPU) 96GB	Ollama	BF16	1.5 GB	fits in fast memory	~286 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M3 Ultra (60c GPU) 96GB	Ollama	BF16	1.5 GB	fits in fast memory	~780 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M3 Ultra (80c GPU) 96GB	Ollama	BF16	1.5 GB	fits in fast memory	~780 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M5 Ultra (64c GPU) 96GB	Ollama	BF16	1.5 GB	fits in fast memory	~1143 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M5 Ultra (80c GPU) 96GB	Ollama	BF16	1.5 GB	fits in fast memory	~1143 tok/s	`ollama run laya:421m-english-bf16`
M2 Ultra 128GB	Ollama	BF16	1.5 GB	fits in fast memory	~762 tok/s	`ollama run laya:421m-english-bf16`
M3 Max (40c GPU) 128GB	Ollama	BF16	1.5 GB	fits in fast memory	~381 tok/s	`ollama run laya:421m-english-bf16`
M4 Max (40c GPU) 128GB	Ollama	BF16	1.5 GB	fits in fast memory	~520 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M4 Max (40c GPU) 128GB	Ollama	BF16	1.5 GB	fits in fast memory	~520 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M5 Max (40c GPU) 128GB	Ollama	BF16	1.5 GB	fits in fast memory	~585 tok/s	`ollama run laya:421m-english-bf16`
DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte)	Ollama	BF16	1.5 GB	fits in fast memory	~244 tok/s	`ollama run laya:421m-english-bf16`
M2 Ultra 192GB	Ollama	BF16	1.5 GB	fits in fast memory	~762 tok/s	`ollama run laya:421m-english-bf16`
2x Mac Studio M5 Max (40c GPU) 128GB	llama.cpp	Q8_0 GGUF	1.1 GB	fits in fast memory	~928 tok/s	`llama-server -hf ggml-org/Laya-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node`
Mac Studio M3 Ultra (60c GPU) 256GB	Ollama	BF16	1.5 GB	fits in fast memory	~780 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M3 Ultra (80c GPU) 256GB	Ollama	BF16	1.5 GB	fits in fast memory	~780 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M5 Ultra (64c GPU) 256GB	Ollama	BF16	1.5 GB	fits in fast memory	~1143 tok/s	`ollama run laya:421m-english-bf16`
Mac Studio M5 Ultra (80c GPU) 256GB	Ollama	BF16	1.5 GB	fits in fast memory	~1143 tok/s	`ollama run laya:421m-english-bf16`
2x DGX Spark 128GB	llama.cpp	Q8_0 GGUF	1.1 GB	fits in fast memory	~387 tok/s	`llama-server -hf ggml-org/Laya-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node`
4x Mac Studio M5 Max (40c GPU) 128GB	llama.cpp	Q8_0 GGUF	1.1 GB	fits in fast memory	~819 tok/s	`llama-server -hf ggml-org/Laya-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node`
2x Mac Studio M5 Ultra (80c GPU) 256GB	llama.cpp	Q8_0 GGUF	1.1 GB	fits in fast memory	~1813 tok/s	`llama-server -hf ggml-org/Laya-GGUF:Q8_0 --rpc node2:50052 -c 8192 # run rpc-server on each other node`
Mac Studio M3 Ultra (80c GPU) 512GB	Ollama	BF16	1.5 GB	fits in fast memory	~780 tok/s	`ollama run laya:421m-english-bf16`
4x DGX Spark 128GB	llama.cpp	Q8_0 GGUF	1.1 GB	fits in fast memory	~341 tok/s	`llama-server -hf ggml-org/Laya-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node`
4x Mac Studio M5 Ultra (80c GPU) 256GB	llama.cpp	Q8_0 GGUF	1.1 GB	fits in fast memory	~1600 tok/s	`llama-server -hf ggml-org/Laya-GGUF:Q8_0 --rpc node2:50052,node3:50052,node4:50052 -c 8192 # run rpc-server on each other node`

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