MODEL / META

Llama 4 Scout

Meta's single-H100 fit requires INT4. All 109B parameters must be stored; 17B active is not the memory number.

Source primaire

Paramètres totaux109B17B active / token

Mémoire estimée~78 GBINT4 MoE

Contexte10M advertisedText + Image

LicenseLlama 4 CommunityMultilingual

01 / IDÉAL POUR

long context
vision
datacenter serving

02 / COMPATIBILITÉ ESTIMÉE

Framework Desktop · Ryzen AI Max+ 395 · 128 GBConfortable NVIDIA DGX Spark · 128 GBConfortable Mac Studio M3 Ultra · 256 GBConfortable

03 / DÉPLOYER AVEC

These are runtime examples, not validated installation recipes. Fit estimates assume quantized files; official repositories may hold full-precision weights. MLX, GGUF and TensorRT require matching conversions or engines.

vLLMAdvancedvllm serve meta-llama/Llama-4-Scout-17B-16E-Instruct --dtype auto

TensorRT-LLMAdvancedtrtllm-serve meta-llama/Llama-4-Scout-17B-16E-Instruct

llama.cppMediumllama-server -m /path/to/verified-model.gguf --port 9931

SGLangAdvancedpython3 -m sglang.launch_server --model-path meta-llama/Llama-4-Scout-17B-16E-Instruct --port 30000

Attention

Meta's single-H100 fit requires INT4. All 109B parameters must be stored; 17B active is not the memory number.