fix cfg merge

tests/conftest.py

@@ -10,7 +10,7 @@ import shutil
 import sys
 import tempfile
 import time
-from pathlib import Path, PosixPath
+from pathlib import Path
 from typing import Generator
 
 import datasets
@@ -432,11 +432,6 @@ def module_temp_dir() -> Generator[str, None, None]:
     shutil.rmtree(_temp_dir)
 
 
-@pytest.fixture(scope="function", autouse=True)
-def unique_triton_cache_dir(temp_dir: str | PosixPath) -> None:
-    os.environ["TRITON_CACHE_DIR"] = str(temp_dir) + "/.triton/cache"
-
-
 @pytest.fixture(scope="function", autouse=True)
 def cleanup_monkeypatches():
     from transformers import Trainer

tests/e2e/multigpu/test_llama.py

@@ -32,7 +32,7 @@ def sft_base_cfg():
     cfg = DictDefault(
         base_model="HuggingFaceTB/SmolLM2-135M",
         tokenizer_config="HuggingFaceTB/SmolLM2-135M",  # this has to be manually set since we haven't done validation
-        sequence_len=2048,
+        sequence_len=1024,
         special_tokens={
             "pad_token": "<|endoftext|>",
         },
@@ -442,7 +442,8 @@ class TestMultiGPULlama:
         self, temp_dir, sft_prepared_dataset_alpaca_cfg, fsdp_state_dict_type
     ):
         # pylint: disable=duplicate-code
-        cfg = DictDefault(
+        cfg = (
+            DictDefault(
                 {
                     "pad_to_sequence_len": True,
                     "num_epochs": 1,
@@ -472,6 +473,7 @@ class TestMultiGPULlama:
                     },
                     "use_tensorboard": True,
                 }
+            )
             | sft_prepared_dataset_alpaca_cfg
         )