updated readme

README.md

@@ -12,18 +12,28 @@ pinned: false
 In this lab, different LLM's were trained through Google Colab. We mainly explored Llama-1B-Instruct, through different datasets, aiming to finetune the model into acting as a psychologist.
 
 Ground models evaluated: 
+
 TinyLlama
-smaller, faster, with around 1B parameters
-not so good for sophisticated answers
+
+    - smaller, faster, with around 1B parameters
+    - not so good for sophisticated answers
+
 Llama3.2 _1B_Instruct
+
 Llama3.2 _3B_Instruct
 
+
+
 Data sets used (from Huggingface)
-mlabonne/FineTome-100k
-wassimm/PsycologyDataset
-samhog/psychology-10k
+
+    - mlabonne/FineTome-100k
+    - wassimm/PsycologyDataset
+    - samhog/psychology-10k
+
+
 
 Evaluation method
+
 Evaluating how well the Fine tuned model works as a psychology assistant
 evaluating simply on different fine-tuned models how the same phrase performs on different fine-tuned models
 
@@ -58,15 +68,11 @@ To improve:
 
 Model centric approach
 
-change r=16 to higher dimension, for more complex LORA matrices, capturing more complex patterns
-
-Using bigger model
-
-Training more epochs
-
-limited due to RAM and time constraint
-
-Change learning rate
+    - change r=16 to higher dimension, for more complex LORA matrices, capturing more complex patterns
+    - Using bigger model
+    - Training more epochs
+    - limited due to RAM and time constraint
+    - Change learning rate