Paraphraser

Sample.py

@@ -42,6 +42,10 @@ def format_qa_entry(qa):
 
 
 def sample_random_entry(dataset='', topic='', model='', n=1):
+    """
+        Sample n (cache_size) random entries from the dataset, topic, model
+
+    """
     if dataset == '': 
         dataset = random.choice(DATASETS)
     
@@ -55,7 +59,7 @@ def sample_random_entry(dataset='', topic='', model='', n=1):
     card_lst = sample_card(dataset, topic, model)
     qa, index = sample_QA_entry(dataset, topic, model)
 
-    display_dict, info_dict = process_for_display(card_lst, qa)
+    display_dict, info_dict = process_for_display(card_lst,  qa)
     info_dict['index'] = index
 
     return display_dict, info_dict
@@ -75,15 +79,10 @@ def process_for_display(card_lst, qa):
 
 
 def select_entry(qa_entry, card_lst):
-    # TODO: Automatically select most relevant criterion.
-    # PLACE HOLDER, RETURN THE WHOEL THING
-
-    # if False:
-    #     return '\n'.join(card_lst[:2])
 
     system_prompt = '''
     Your task is to effectively condense the essential details from the student's evaluation card that are most relevant to predicting the correctness of their answer to a question.
-    Limit your paraphrase to 100-150 words, focusing on distilling the key observations and outcomes that are directly pertinent to the inquiry. 
+    Limit your paraphrase to 50-100 words, focusing on distilling the key observations and outcomes that are directly pertinent to the inquiry. 
     It's crucial to present an informative, unbiased summary that retains the integrity of the original card's information. 
     Your goal is to craft a paraphrase that enhances the user's ability to accurately gauge the student's response, by emphasizing relevant insights and conclusions without altering the core facts.
     '''
@@ -96,18 +95,19 @@ def select_entry(qa_entry, card_lst):
         ## Evaluation Card:
         {card_str}
 
-        Again, your task is not to answer the question, but summarize the student's ability in answering the question!
+        Again, your task is not to answer the question, but summarize the student's ability in answering the question! Only 100 words max! Use bullet points.
+        Only relevant information to the question is needed.
     '''
 
-    # Mistral-7B-Instruct-v0.2
-    # mistralai/Mixtral-8x7B-Instruct-v0.1
+    model_avaliable = {'mixtral': 'mistralai/Mixtral-8x7B-Instruct-v0.1',
+                       'mistral': 'mistralai/Mistral-7B-Instruct-v0.2'}
 
     model = HFAPIModel(system_prompt=system_prompt,
-                        model_name='mistralai/Mistral-7B-Instruct-v0.2')
+                        model_name=model_avaliable['mistral'])
 
     
-    response = model(prompt)
-    print(response)
+    response = model(prompt).replace('\n', '\n\n')
+
     del model
     return response