hi, Million (
@milliondotjs) has $1.3M in gpu credits that expires in a year.
we are looking to fund experiments for:
- determining the most optimal training curriculum, reward modeler, or model merging combination with evolutionary algorithms (any domain ok)
- a diffusion text decoder (for a text-JEPA)
- training a model to use theorem provers (lean, isabelle) with proofs/tactics search to solve IMO problems (what's the next domain after LeanDojo, FunSearch, AlphaGeometry? how to improve domain/proof-specific tactics search?)
- compressing a learned functions library with llms (LILO generated docs, how to create higher order abstractions guided by a deterministic graph ranker?)
- using GFlowNet to improve llm reasoning, with a focus on code optimization problems (ref:
@edwardjhu "Amortizing Intractable Inference in LLMs")
- scaling energy transformers (ref:
@Ben_Hoov)
please DM or email john @ million . dev (
@johnjyang)
if these questions interest you, Million is also hiring for 1-2 talented ML engineers! we are focused on code optimization, but broadly curious about topics in swe and theorem proving automations.
Open to grants, part-time or full-time work