tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Fix: shorten mem logs to 2 decimal places and renamed nd docs (#3011) [skip ci]

Browse Source 
* fix: shorten memory logs

* fix: title name
This commit is contained in:
NanoCode012
2025-08-04 21:23:36 +07:00
committed by GitHub
parent 5691992d34
commit a54c1be972
2 changed files with 7 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
docs/nd_parallelism.qmd
View File

@@ -1,4 +1,6 @@
# N-D Parallelism ---
title: "N-D Parallelism"
---
Axolotl enables training models at scale by composing different parallelism techniques. This is essential when: Axolotl enables training models at scale by composing different parallelism techniques. This is essential when:

8
src/axolotl/core/trainers/base.py
View File

@@ -567,10 +567,10 @@ class AxolotlTrainer(
# Add memory usage # Add memory usage
try: try:
active, allocated, reserved = get_gpu_memory_usage() active, allocated, reserved = get_gpu_memory_usage()
logs["memory/max_memory_active"] = active logs["memory/max_memory_active(gib)"] = round(active, 2)
logs["memory/max_memory_allocated"] = allocated logs["memory/max_memory_allocated(gib)"] = round(allocated, 2)
logs["memory/device_memory_reserved"] = reserved logs["memory/device_memory_reserved(gib)"] = round(reserved, 2)
except (ValueError, FileNotFoundError): except (ValueError, TypeError, FileNotFoundError):
pass pass
del self._stored_metrics[train_eval] del self._stored_metrics[train_eval]