Tesla V100 32GB Llama-3.1-8B Q4_K_M decode (homelab field guide claim)
60.0 tok/s CLAIMED ('60+' on Pop!_OS); vramguide models 106 - treat 60-100 as the envelope
Fields
id
benchmarks/comm-v100-32gb-llama3-8b-q4-decode-field-guideMetric
decode_tok_s
Value
60
Unit
tok/s CLAIMED ('60+' on Pop!_OS); vramguide models 106 - treat 60-100 as the envelope
Model
Llama-3.1-8B Q4_K_M
Methodology
HF 'V100 Homelab Field Guide' claimed measured figure; not independently reproduced here
Measured by
third_party
Thermal state
unknown