tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

use smaller pretrained models for ci (#3620) [skip ci]

Browse Source 
* use smaller pretrained models for ci

* more steps for loss check

* fix tests

* more train steps

* fix losses
This commit is contained in:
Wing Lian
2026-04-27 13:22:56 -04:00
committed by GitHub
parent 798c8fba89
commit ac77da96da
24 changed files with 716 additions and 288 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/e2e/solo/test_batch_flattening.py
View File

@@ -18,7 +18,7 @@ from transformers import AutoModelForCausalLM
# Import the actual trainer methods we want to test
from axolotl.core.trainers.grpo.async_trainer import AsyncGRPOTrainer
MODEL_NAME = "Qwen/Qwen3-0.6B"
MODEL_NAME = "axolotl-ai-co/tiny-qwen3-129m"
def _fix_patched_attention(model):