Stanford UniversityAI Measurement ScienceAIMS — AI Measurement Science

General

MorphKV (LongGenBench)

LongGenBench long-response constraint-following outcomes for three KV-cache compression methods (MorphKV / SnapKV / H2O) across three base LLMs; response is per-constraint yes/no.

40items
9subjects
MITlicense
reasoningdomain
textmodality
item-level responses released
Saturation status: Unknown
Original sourcePaper← All benchmarks

Response matrix

Fit to width. Hover for subject & item; click a cell for details.

MorphKV (LongGenBench) response matrix: AI models (rows) against items (columns)
Correct (1)Incorrect (0)Unobserved

Scale: 1 = correct · 0 = incorrect

Subjects

  1. 1Mistral-7B-Instruct-v0.2 + H2O0.447
  2. 2Mistral-7B-Instruct-v0.2 + SnapKV0.4343
  3. 3Qwen2.5-7B-Instruct + MorphKV0.4302
  4. 4Mistral-7B-Instruct-v0.2 + MorphKV0.4286
  5. 5Llama-3.1-8B-Instruct + SnapKV0.4093
  6. 6Llama-3.1-8B-Instruct + H2O0.3977
  7. 7Llama-3.1-8B-Instruct + MorphKV0.3857
  8. 8Qwen2.5-7B-Instruct + H2O0.3813
  9. 9Qwen2.5-7B-Instruct + SnapKV0.3767
Full data on Hugging FaceBack to the gallery