Release training code

0bf5e500 · Tri Dao · 9bc63d1e · 0bf5e500 · 0bf5e500 · 0bf5e500
Commit 0bf5e500 authored Nov 28, 2022 by Tri Dao
20 changed files
--- a/training/run.py
+++ b/training/run.py
--- a/training/src/callbacks/__init__.py
+++ b/training/src/callbacks/__init__.py
--- a/training/src/callbacks/causality_monitor.py
+++ b/training/src/callbacks/causality_monitor.py
--- a/training/src/callbacks/ema.py
+++ b/training/src/callbacks/ema.py
--- a/training/src/callbacks/flop_count.py
+++ b/training/src/callbacks/flop_count.py
--- a/training/src/callbacks/gpu_affinity.py
+++ b/training/src/callbacks/gpu_affinity.py
--- a/training/src/callbacks/loss_scale_monitor.py
+++ b/training/src/callbacks/loss_scale_monitor.py
--- a/training/src/callbacks/model_checkpoint.py
+++ b/training/src/callbacks/model_checkpoint.py
--- a/training/src/callbacks/norm_monitor.py
+++ b/training/src/callbacks/norm_monitor.py
--- a/training/src/callbacks/params_log.py
+++ b/training/src/callbacks/params_log.py
--- a/training/src/callbacks/speed_monitor.py
+++ b/training/src/callbacks/speed_monitor.py
--- a/training/src/callbacks/wandb_callbacks.py
+++ b/training/src/callbacks/wandb_callbacks.py
--- a/training/src/datamodules/datasets/detokenizer.py
+++ b/training/src/datamodules/datasets/detokenizer.py
--- a/training/src/datamodules/datasets/lm_dataset.py
+++ b/training/src/datamodules/datasets/lm_dataset.py
--- a/training/src/datamodules/fault_tolerant_sampler.py
+++ b/training/src/datamodules/fault_tolerant_sampler.py
--- a/training/src/datamodules/imagenet.py
+++ b/training/src/datamodules/imagenet.py
--- a/training/src/datamodules/language_modeling_hf.py
+++ b/training/src/datamodules/language_modeling_hf.py
--- a/training/src/datamodules/timm_mixup.py
+++ b/training/src/datamodules/timm_mixup.py
--- a/training/src/distributed/ddp_comm_hooks.py
+++ b/training/src/distributed/ddp_comm_hooks.py
--- a/training/src/eval.py
+++ b/training/src/eval.py