Release: 0.0.4

setup.py

@@ -57,7 +57,7 @@
 # Setup
 setup(
     name="xllm",
-    version="0.0.3-dev",
+    version="0.0.4",
     description="Simple & Cutting Edge LLM Finetuning",
     license_files=["LICENSE"],
     long_description=open("README.md", "r", encoding="utf-8").read(),

src/xllm/__init__.py

@@ -14,7 +14,7 @@
 
 # ruff: noqa: F401
 
-__version__ = "0.0.3-dev"
+__version__ = "0.0.4"
 
 from . import enums, types
 from .cli.fuse import cli_run_fuse

src/xllm/experiments/base.py

@@ -119,11 +119,13 @@ def build(self):
             self.before_eval_dataset_build()
             self.eval_dataset = self.build_eval_dataset()
             if self.eval_dataset is not None:
+                if self.training_arguments is not None:
+                    self.training_arguments.do_eval = True
                 dist_logger(
-                    f"Valid dataset {self.eval_dataset.__class__.__name__} was built. Size: {len(self.eval_dataset)}"
+                    f"Eval dataset {self.eval_dataset.__class__.__name__} was built. Size: {len(self.eval_dataset)}"
                 )
             else:
-                dist_logger("Valid dataset is None")
+                dist_logger("Eval dataset is None")
             self.after_eval_dataset_build()
 
         if self.tokenizer is None:

src/xllm/utils/post_training.py

@@ -72,7 +72,7 @@ def fuse_lora(config: HuggingFaceConfig) -> Tuple[PreTrainedTokenizer, PreTraine
                 json.dump(tokenizer_config, file_object, indent=2)
         logger.info(f"Model saved locally to {config.fused_model_local_path}")
 
-    if config.push_to_hub:
+    if config.push_to_hub or config.hub_model_id is not None:
         logger.info(f"Pushing model to the hub {config.hub_model_id}")
         if config.hub_model_id is not None:
             tokenizer.push_to_hub(