L4

18 records across 2 models (quants: BF16, PQ2_0, PTQ1_0, Ternary PQ2_0, Ternary PTQ1_0).

hardwaremodelquantbackendtok/spp tok/sWttft sctx/tgdateclassflagssource
L4Qwen3 4B BF16vLLM82.8––––2026-10-01community–source
L4Qwen3 4B BF16vLLM81.6––––2026-10-01community–source
L4Qwen3 4B BF16vLLM80.5––––2026-10-01community–source
L4Qwen3 4B BF16vLLM79.7––––2026-10-01community–source
L4Qwen3 4B BF16vLLM78.2––––2026-10-01community–source
L4Qwen3 4B BF16vLLM77.0––––2026-10-01community–source
L4Qwen3 4B BF16vLLM61.2––––2026-10-01community–source
L4Qwen3 4B BF16vLLM39.2––––2026-10-01communitycontradictionsource
L4Ternary-Bonsai-2-27B PQ2_0llama.cpp (PrismML fork)68.4––––2026-09community–source
L4Ternary-Bonsai-2-27B PQ2_0llama.cpp (PrismML fork)67.8––––2026-09community–source
L4Ternary-Bonsai-2-27B PQ2_0llama.cpp (PrismML fork)67.6––––2026-09community–source
L4Ternary-Bonsai-2-27B PQ2_0llama.cpp (PrismML fork)42.9––––2026-09community–source
L4Ternary-Bonsai-2-27B Ternary PTQ1_0llama.cpp (CUDA)32.1467––tg1282026-09-16community–source
L4Ternary-Bonsai-2-27B PTQ1_0llama.cpp (PrismML fork)31.6––––2026-09community–source
L4Ternary-Bonsai-2-27B PTQ1_0llama.cpp (PrismML fork)31.4––––2026-09community–source
L4Ternary-Bonsai-2-27B PTQ1_0llama.cpp (PrismML fork)31.3––––2026-09community–source
L4Ternary-Bonsai-2-27B PTQ1_0llama.cpp (PrismML fork)30.6––––2026-09community–source
L4Ternary-Bonsai-2-27B Ternary PQ2_0llama.cpp (CUDA)29.8777––tg1282026-09-16community–source

All rows cite their source. Provenance: community 40018 sourced 1085

← all hardware