Search results

Why step+1 twice in run_training_loop?

Status: Open.

#10 In mit-han-lab/fastrl;

· Weigaa opened on Feb 3, 2026

How to run ngram speculative decoding.

Status: Open.

#8 In mit-han-lab/fastrl;

· zzhbrr opened on Dec 18, 2025

Significant Performance Degradation with Speculative Decoding on H20 GPUs

Status: Open.

#7 In mit-han-lab/fastrl;

· zzhbrr opened on Dec 18, 2025

Unexpected long reward time in example script

Status: Open.

#6 In mit-han-lab/fastrl;

· baihuajun24 opened on Dec 16, 2025

No acceleration in GRPO training with Eagle1 speculative decoding

Status: Open.

#5 In mit-han-lab/fastrl;

· Jikai0Wang opened on Dec 10, 2025

I want to train my own model, so do I need to first train an Eagle model for it?

Status: Open.

#1 In mit-han-lab/fastrl;

· zwc163 opened on Nov 25, 2025