Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -1,35 +1,45 @@
|
|
1 |
-
from transformers import AutoTokenizer,
|
2 |
from peft import PeftModel, PeftConfig
|
3 |
-
|
4 |
-
|
5 |
import gradio as gr
|
6 |
|
7 |
# Use the base model's ID
|
8 |
base_model_id = "mistralai/Mistral-7B-v0.1"
|
9 |
model_directory = "Tonic/mistralmed"
|
10 |
|
11 |
-
#
|
12 |
-
|
13 |
-
config = PeftConfig.from_pretrained("Tonic/mistralmed", token="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")
|
14 |
-
model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1")
|
15 |
-
model = PeftModel.from_pretrained(model, "Tonic/mistralmed", token="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")
|
16 |
tokenizer = AutoTokenizer.from_pretrained(base_model_id, trust_remote_code=True)
|
17 |
tokenizer.pad_token = tokenizer.eos_token
|
18 |
tokenizer.padding_side = 'left'
|
19 |
|
|
|
|
|
|
|
|
|
|
|
20 |
class ChatBot:
|
21 |
def __init__(self):
|
22 |
self.history = []
|
23 |
|
24 |
def predict(self, input):
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
bot = ChatBot()
|
35 |
|
|
|
1 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
2 |
from peft import PeftModel, PeftConfig
|
3 |
+
import torch
|
|
|
4 |
import gradio as gr
|
5 |
|
6 |
# Use the base model's ID
|
7 |
base_model_id = "mistralai/Mistral-7B-v0.1"
|
8 |
model_directory = "Tonic/mistralmed"
|
9 |
|
10 |
+
# Instantiate the Models
|
|
|
|
|
|
|
|
|
11 |
tokenizer = AutoTokenizer.from_pretrained(base_model_id, trust_remote_code=True)
|
12 |
tokenizer.pad_token = tokenizer.eos_token
|
13 |
tokenizer.padding_side = 'left'
|
14 |
|
15 |
+
# Load the PEFT model
|
16 |
+
peft_config = PeftConfig.from_pretrained("Tonic/mistralmed")
|
17 |
+
base_model = AutoModelForSeq2SeqLM.from_pretrained(model_directory)
|
18 |
+
peft_model = PeftModel.from_pretrained(base_model, "Tonic/mistralmed")
|
19 |
+
|
20 |
class ChatBot:
|
21 |
def __init__(self):
|
22 |
self.history = []
|
23 |
|
24 |
def predict(self, input):
|
25 |
+
# Encode user input
|
26 |
+
user_input_ids = tokenizer.encode(input + tokenizer.eos_token, return_tensors="pt")
|
27 |
+
|
28 |
+
# Concatenate the user input with chat history
|
29 |
+
if self.history:
|
30 |
+
chat_history_ids = torch.cat([self.history, user_input_ids], dim=-1)
|
31 |
+
else:
|
32 |
+
chat_history_ids = user_input_ids
|
33 |
+
|
34 |
+
# Generate a response using the PEFT model
|
35 |
+
response = peft_model.generate(chat_history_ids, max_length=512, pad_token_id=tokenizer.eos_token_id)
|
36 |
+
|
37 |
+
# Update chat history
|
38 |
+
self.history = response
|
39 |
+
|
40 |
+
# Decode and return the response
|
41 |
+
response_text = tokenizer.decode(response[0], skip_special_tokens=True)
|
42 |
+
return response_text
|
43 |
|
44 |
bot = ChatBot()
|
45 |
|