Update README.md

README.md

@@ -1,3 +1,52 @@
 ---
 license: mit
 ---
+
+how to use model with gpt4all
+
+
+import pandas as pd
+from datasets import load_dataset
+import re
+csv_file = "/Users/admin/Downloads/GPT4_output_GPT-2_training_df.csv"
+# Load the dataset
+dataset = load_dataset("csv", data_files=csv_file, split="train")
+
+def stop_on_token_callback(token_id, token_string):
+    # one sentence is enough:
+    if '<eos>' in token_string:
+        return False
+    else:
+        return True
+
+alpaca_prompt = """Below is an instruction that describes a task, paired with an 
+input that provides further context. Write a response that appropriately completes 
+the request.
+### Instruction:
+Find out which character said this specific quote along with their gender
+### Input:
+{}
+### Response:
+{}"""
+num_examples=10
+from gpt4all import GPT4All
+model = GPT4All('/Users/admin/Downloads/model-unsloth.Q4_K_M.gguf')
+system_template = '''Below is an instruction that describes a task, paired with an 
+input that provides further context. Write a response that appropriately completes 
+the request.'''
+# many models use triple hash '###' for keywords, Vicunas are simpler:
+prompt_template = """### Instruction:
+Find out which character said this specific quote along with their gender
+### Input:
+{0}
+### Response:
+"""
+with model.chat_session(system_template, prompt_template):
+    for i in range(min(num_examples, len(dataset))):
+        row = dataset[i]
+        #response1 = model.generate(row['formatted_input'], callback=stop_on_token_callback)
+        print(i)
+        response1 = model.generate(row['formatted_input'])
+        print(response1)
+        print()
+        print(f"Correct output: {row['TrueSpeaker']}")