little faster generation + prints ot a examplke generatino in validation mode, more optimization in trianing function

2025-12-16 02:08:55 +08:00 · 2025-02-21 16:02:27 +01:00
parent 11c8991476
commit 2f20107d9b
2 changed files with 119 additions and 59 deletions
--- a/llms/mlx_lm/lora.py
+++ b/llms/mlx_lm/lora.py
@@ -374,7 +374,9 @@ def evaluate_model(args, model: nn.Module, tokenizer: TokenizerWrapper, test_set
            max_seq_length=args.max_seq_length,
            beta=args.beta,
            group_size=args.group_size,
-            epsilon=args.epsilon
+            epsilon=args.epsilon,
+            temperature=args.temperature,
+            max_tokens=args.max_seq_length
        )

        test_ppl = math.exp(test_loss)