eval_framework / docs /EXPERIMENTS.md
LCZZZZ's picture
Upload eval_framework source code
85b19cf verified

实验记录

实验环境

  • 数据:VisTrajQA VAB smoke(1 sample, vab_minecraft, 30 步, 5 sessions, 45 QA)
  • 转换模式:text-only(image caption 内联到 user turn 文本)
  • Judge / Answer 模型:从 .env 读取 OPENAI_MODEL
  • 日期:2026-04-15

总表

Memory 维度

Baseline Recall ↑ Update Recall ↑ Correctness ↑ Hallucination ↓ Irrelevant ↓ Update Score ↑
Dummy 95.0% 90.0% 50.0% 5.0% 45.0% 71.4%
FUMemory 95.0% 90.0% 80.0% 10.0% 10.0% 69.0%
STMemory 95.0% 90.0% 83.3% 6.7% 10.0% 71.4%
LTMemory 95.0% 90.0% 83.3% 6.7% 10.0% 69.0%
GAMemory 92.1% 90.0% 62.5% 22.5% 15.0% 71.4%
MGMemory 95.0% 90.0% 83.3% 6.7% 10.0% 71.4%
RFMemory 95.0% 90.0% 90.0% 6.7% 3.3% 71.4%
MMMemory 95.0% 90.0% 83.3% 6.7% 10.0% 71.4%
MMFUMemory 95.0% 90.0% 83.3% 6.7% 10.0% 71.4%
NGMemory 95.0% 90.0% 83.3% 6.7% 10.0% 71.4%
AUGUSTUSMemory 95.0% 90.0% 83.3% 6.7% 10.0% 71.4%
UniversalRAGMemory 95.0% 90.0% 80.0% 10.0% 10.0% 71.4%
Mem0 47.9% 56.7% 57.3% 18.0% 24.7% 81.0%
Mem0-Graph 48.2% 56.7% 37.9% 14.6% 38.1% 71.4%
SimpleMem 95.0% 90.0% 48.3% 5.0% 46.7% 71.4%
Omni-SimpleMem
MemVerse 95.0% 90.0% 65.0% 23.3% 11.7% 71.4%
Zep
A-Mem 95.0% 90.0% 56.7% 5.0% 38.3% 71.4%
MemoryOS 47.1% 41.0% 33.8% 3.5% 62.8% 50.0%

QA 维度

Baseline QA Correct ↑ QA Hallucination ↓ QA Omission ↓ Evidence Coverage ↑
Dummy 44.4% 24.4% 31.1% 47.6%
FUMemory 57.8% 40.0% 2.2% 65.0%
STMemory 31.1% 33.3% 35.6% 32.0%
LTMemory 40.0% 26.7% 33.3% 51.5%
GAMemory 37.8% 20.0% 42.2% 51.5%
MGMemory 60.0% 35.6% 4.4% 68.9%
RFMemory 60.0% 37.8% 2.2% 63.1%
MMMemory 57.8% 40.0% 2.2% 66.0%
MMFUMemory 60.0% 37.8% 2.2% 67.0%
NGMemory 62.2% 24.4% 13.3% 71.8%
AUGUSTUSMemory 64.4% 22.2% 13.3% 68.9%
UniversalRAGMemory 40.0% 28.9% 31.1% 49.5%
Mem0 26.7% 28.9% 44.4% 31.1%
Mem0-Graph 31.1% 20.0% 48.9% 32.0%
SimpleMem 57.8% 22.2% 20.0% 35.9%
Omni-SimpleMem
MemVerse 40.0% 28.9% 31.1% 40.8%
Zep
A-Mem 46.7% 20.0% 33.3% 50.5%
MemoryOS 28.9% 24.4% 46.7% 37.9%

QA 分类型正确率

Baseline FR DU MB TR KR VFR VS VU CMR
Dummy 5/5 0/5 4/5 2/5 2/5 4/5 0/5 1/5 2/5
FUMemory 4/5 3/5 5/5 4/5 4/5 4/5 0/5 2/5 0/5
STMemory 0/5 0/5 5/5 1/5 3/5 2/5 0/5 0/5 3/5
LTMemory 2/5 1/5 5/5 1/5 3/5 4/5 0/5 0/5 2/5
GAMemory 4/5 0/5 5/5 0/5 2/5 4/5 0/5 0/5 2/5
MGMemory 4/5 3/5 5/5 4/5 4/5 4/5 0/5 1/5 2/5
RFMemory 4/5 3/5 5/5 4/5 4/5 4/5 0/5 1/5 2/5
MMMemory 4/5 3/5 5/5 4/5 4/5 4/5 0/5 2/5 0/5
MMFUMemory 4/5 3/5 5/5 4/5 4/5 4/5 0/5 2/5 1/5
NGMemory 5/5 3/5 5/5 3/5 3/5 4/5 0/5 2/5 3/5
AUGUSTUSMemory 5/5 2/5 5/5 4/5 3/5 4/5 1/5 2/5 3/5
UniversalRAGMemory 2/5 1/5 5/5 1/5 3/5 4/5 0/5 0/5 2/5
Mem0 2/5 1/5 5/5 0/5 0/5 2/5 0/5 0/5 2/5
Mem0-Graph 2/5 0/5 5/5 0/5 1/5 3/5 0/5 0/5 3/5
SimpleMem 5/5 3/5 4/5 3/5 2/5 2/5 0/5 3/5 4/5
Omni-SimpleMem
MemVerse 5/5 0/5 5/5 0/5 2/5 3/5 0/5 0/5 3/5
Zep
A-Mem 3/5 3/5 5/5 0/5 2/5 4/5 0/5 0/5 4/5
MemoryOS 4/5 1/5 5/5 0/5 0/5 1/5 0/5 0/5 2/5