BREAKING
Renmin, Tsinghua drop DeepAnalyze-8B
Raw data to full report, end to end
1
Prepare
↓
2
Analyze
↓
3
Model
↓
4
Visualize
↓
5
Report
0
B
parameters
0
benchmarks
0
K
training set
Trainable agent vs workflow control
Old approach
Workflow
●
Steers GPT-4 style models
●
Predefined steps
●
Struggles with multi-step
DeepAnalyze
Agentic
●
Curriculum training
●
Trajectory synthesis
●
Adaptive optimization
Fully open: model, code, data
CoDA-Bench and DeepPrep next
AI NEWS BLITZ
Renmin University and Tsinghua have released an 8B open-source data-analysis agent.