PEFT
English
medical
Tonic commited on
Commit
933497b
1 Parent(s): e0163b4

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +58 -9
README.md CHANGED
@@ -67,19 +67,68 @@ Use the code below to get started with the model.
67
  [Tonic/MistralMED_Chat](https://huggingface.co/Tonic/MistralMED_Chat)
68
 
69
  ```python
70
- from transformers import AutoConfig, AutoTokenizer, AutoModelForSeq2SeqLM, AutoModelForCausalLM, MistralForCausalLM
71
- from peft import PeftModel, PeftConfig
72
  import torch
73
  import gradio as gr
74
-
75
-
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
76
  # Use the base model's ID
77
  base_model_id = "mistralai/Mistral-7B-v0.1"
78
  model_directory = "Tonic/mistralmed"
79
 
80
- # Instantiate the Models
81
  tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1", trust_remote_code=True, padding_side="left")
82
- #tokenizer.pad_token = tokenizer.eos_token
 
83
  tokenizer.padding_side = 'left'
84
 
85
 
@@ -98,9 +147,9 @@ class ChatBot:
98
  def __init__(self):
99
  self.history = []
100
 
101
- def predict(self, input):
102
  # Encode user input
103
- user_input_ids = tokenizer.encode(input + tokenizer.eos_token, return_tensors="pt")
104
 
105
  # Concatenate the user input with chat history
106
  if len(self.history) > 0:
@@ -122,7 +171,7 @@ bot = ChatBot()
122
 
123
  title = "👋🏻Welcome to Tonic's MistralMed Chat🚀"
124
  description = "You can use this Space to test out the current model (MistralMed) or duplicate this Space and use it for any other model on 🤗HuggingFace. Join me on Discord to build together."
125
- examples = [["What is the boiling point of nitrogen"]]
126
 
127
  iface = gr.Interface(
128
  fn=bot.predict,
 
67
  [Tonic/MistralMED_Chat](https://huggingface.co/Tonic/MistralMED_Chat)
68
 
69
  ```python
70
+ from transformers import AutoTokenizer, MistralForCausalLM
 
71
  import torch
72
  import gradio as gr
73
+ import random
74
+ from textwrap import wrap
75
+ import random
76
+
77
+ # Functions to Wrap the Prompt Correctly
78
+
79
+ def wrap_text(text, width=90):
80
+ lines = text.split('\n')
81
+ wrapped_lines = [textwrap.fill(line, width=width) for line in lines]
82
+ wrapped_text = '\n'.join(wrapped_lines)
83
+ return wrapped_text
84
+
85
+ def multimodal_prompt(input_text, system_prompt="", max_length=512):
86
+ """
87
+ Generates text using a large language model, given a prompt and a device.
88
+ Args:
89
+ input_text: The input text to generate a response for.
90
+ system_prompt: Optional system prompt.
91
+ max_length: Maximum length of the generated text.
92
+ Returns:
93
+ A string containing the generated text.
94
+ """
95
+ # Modify the input text to include the desired format
96
+ formatted_input = f"""<s>[INST]{input_text}[/INST]"""
97
+
98
+ # Encode the input text
99
+ encodeds = tokenizer(formatted_input, return_tensors="pt", add_special_tokens=False)
100
+ model_inputs = encodeds.to(device)
101
+
102
+ # Generate a response using the model
103
+ output = model.generate(
104
+ **model_inputs,
105
+ max_length=max_length,
106
+ use_cache=True,
107
+ early_stopping=True,
108
+ bos_token_id=model.config.bos_token_id,
109
+ eos_token_id=model.config.eos_token_id,
110
+ pad_token_id=model.config.eos_token_id,
111
+ temperature=0.1,
112
+ do_sample=True
113
+ )
114
+
115
+ # Decode the response
116
+ response_text = tokenizer.decode(output[0], skip_special_tokens=True)
117
+
118
+ return response_text
119
+
120
+
121
+ # Define the device
122
+ device = "cuda" if torch.cuda.is_available() else "cpu"
123
+
124
  # Use the base model's ID
125
  base_model_id = "mistralai/Mistral-7B-v0.1"
126
  model_directory = "Tonic/mistralmed"
127
 
128
+ # Instantiate the Tokenizer
129
  tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1", trust_remote_code=True, padding_side="left")
130
+ # tokenizer = AutoTokenizer.from_pretrained("Tonic/mistralmed", trust_remote_code=True, padding_side="left")
131
+ tokenizer.pad_token = tokenizer.eos_token
132
  tokenizer.padding_side = 'left'
133
 
134
 
 
147
  def __init__(self):
148
  self.history = []
149
 
150
+ def predict(self, input_text):
151
  # Encode user input
152
+ user_input_ids = tokenizer.encode(input_text, return_tensors="pt")
153
 
154
  # Concatenate the user input with chat history
155
  if len(self.history) > 0:
 
171
 
172
  title = "👋🏻Welcome to Tonic's MistralMed Chat🚀"
173
  description = "You can use this Space to test out the current model (MistralMed) or duplicate this Space and use it for any other model on 🤗HuggingFace. Join me on Discord to build together."
174
+ examples = [["What is the boiling point of nitrogen?"]]
175
 
176
  iface = gr.Interface(
177
  fn=bot.predict,