Update README.md

README.md

@@ -1,9 +1,8 @@
 ---
-base_model: unsloth/Qwen2.5-3B-Instruct
+base_model: Qwen2.5-3B-Instruct
 tags:
 - text-generation-inference
 - transformers
-- unsloth
 - qwen2
 - trl
 - sft
@@ -53,5 +52,8 @@ response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
 ```
 
 # Notes:
-- Focus on o and ignore all linear layer
+- For small datasets with narrow content which the model already has well, and doesn't want the model to forget the knowledge by focusing on o. 
 - Fine-tuned lora with rank = 16 and alpha = 32, epoch = 1
+
+# Improvement
+- Increasing rank can help the model do better at your robust structure.