Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
LLM Pretraining on MLPerf C4
Loading...
20
Train step throughput
MXFP4 + H16
19
19.5
20
20.5
May 11, 2026
Train step throughput
Convergence Overhead (Tokens)
End-to-end Speedup
Updated 21d ago
Evaluation Results
Method
Method
Links
Train step throughput
Convergence Overhead (Tokens)
End-to-end Speedup
MXFP4 + H16
Backbone=Llama 3.1-8B,...
2026.05
20
8
9
Feedback
Search any
task
Search any
task
