List your company

Inference Performance Evaluation on GPU Context Length Sweep (2K-8K)

116TTFTSpectrumKV
54.08472.048901,307.96Jun 7, 2026

Evaluation results

MethodLinks
SpectrumKVBackbone=Mistral-7B, Sequence length=2K, KV Cache Budget (b)=0.52026.061165850.13
SpectrumKVBackbone=Gemma-2-9B, Sequence length=2K, KV Cache Budget (b)=0.52026.061355038.92
SpectrumKVBackbone=Qwen2.5-7B, Sequence length=2K, KV Cache Budget (b)=0.52026.061546143.511
SpectrumKVBackbone=Mistral-7B, Sequence length=4K, KV Cache Budget (b)=0.52026.062335349.14
Gemma-2-9BSequence length=2K, KV Cache Budget (b)=Full2026.06269—38.2—
SpectrumKVBackbone=Gemma-2-9B, Sequence length=4K, KV Cache Budget (b)=0.52026.062706238.233
Mistral-7BSequence length=2K, KV Cache Budget (b)=Full2026.06274—51.9—
SpectrumKVBackbone=Qwen2.5-7B, Sequence length=4K, KV Cache Budget (b)=0.52026.062965343.22
Qwen2.5-7BSequence length=2K, KV Cache Budget (b)=Full2026.06391—49.2—
Mistral-7BSequence length=4K, KV Cache Budget (b)=Full2026.06497—47.1—
SpectrumKVBackbone=Mistral-7B, Sequence length=8K, KV Cache Budget (b)=0.52026.0650259477
Qwen2.5-7BSequence length=4K, KV Cache Budget (b)=Full2026.06630—42.4—
Gemma-2-9BSequence length=4K, KV Cache Budget (b)=Full2026.06709—28.8—
SpectrumKVBackbone=Gemma-2-9B, Sequence length=8K, KV Cache Budget (b)=0.52026.067135728.85
Mistral-7BSequence length=8K, KV Cache Budget (b)=Full2026.061,211—44.1—
Gemma-2-9BSequence length=8K, KV Cache Budget (b)=Full2026.061,664—27.3—