flexflow / FlexFlow

FlexFlow Serve: Low-Latency, High-Performance LLM Serving
https://flexflow.readthedocs.io
Apache License 2.0
1.67k stars 224 forks source link

Separate up DP logic in compiler #1325

Open wmdi opened 6 months ago

wmdi commented 6 months ago

Separate the problem logic (e.g., dp states) from the search logic (dynamic programming), and separate the exploration logic from the evaluation logic. Need to consider how to design the interfaces.