fix(hf): pass fake eval dataset since it is required (#853)

iterative · Dec 19, 2024 · dac96b2 · dac96b2
1 parent a15bb31
commit dac96b2
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 1 deletion.
diff --git a/src/dvclive/huggingface.py b/src/dvclive/huggingface.py
@@ -73,7 +73,10 @@ def on_train_end(
     ):
         if self._log_model is True and state.is_world_process_zero:
             fake_trainer = Trainer(
-                args=args, model=kwargs.get("model"), tokenizer=kwargs.get("tokenizer")
+                args=args,
+                model=kwargs.get("model"),
+                tokenizer=kwargs.get("tokenizer"),
+                eval_dataset=["fake"],
             )
             name = "best" if args.load_best_model_at_end else "last"
             output_dir = os.path.join(args.output_dir, name)

diff --git a/tests/frameworks/test_huggingface.py b/tests/frameworks/test_huggingface.py
@@ -162,6 +162,7 @@ def test_huggingface_log_model(
         live_callback = callback(live=live, log_model=log_model)
 
     args.load_best_model_at_end = best
+    args.metric_for_best_model = "loss"
 
     trainer = Trainer(
         model,