Commit
•
e66fc88
1
Parent(s):
41f3c42
Update README.md
Browse files
README.md
CHANGED
@@ -38,6 +38,95 @@ While VergilGPT2 showcases impressive conversational capabilities, it is importa
|
|
38 |
|
39 |
Unleash the power of VergilGPT2 and immerse yourself in dynamic and captivating conversations that push the boundaries of interactive AI experiences. Let the conversations unfold and discover the true potential of this extraordinary model.
|
40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
## Installation
|
42 |
|
43 |
Make sure to install the required dependencies by running the following commands:
|
|
|
38 |
|
39 |
Unleash the power of VergilGPT2 and immerse yourself in dynamic and captivating conversations that push the boundaries of interactive AI experiences. Let the conversations unfold and discover the true potential of this extraordinary model.
|
40 |
|
41 |
+
## Engaging with Vergil
|
42 |
+
If you're eager to have a conversation with VergilGPT2, you can utilize the following code snippet. Feel free to experiment with the variables 'temperature', 'top_k', 'top_p', to customize the response generation according to your preferences.
|
43 |
+
|
44 |
+
```python
|
45 |
+
import torch
|
46 |
+
import re
|
47 |
+
import random
|
48 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
49 |
+
|
50 |
+
def filter_english_words(text):
|
51 |
+
# Initialize an empty list to store the filtered words
|
52 |
+
filtered_text = []
|
53 |
+
# Split the input text into individual words
|
54 |
+
words = text.split()
|
55 |
+
# Iterate through the words
|
56 |
+
for word in words:
|
57 |
+
# Check if the word is a valid English word or a punctuation
|
58 |
+
if word.isalpha() or re.match(r'^[!"#$%&\'()*+,-./:;<=>?@[\]^_`{|}~]$', word):
|
59 |
+
# If it is, append it to the filtered text list
|
60 |
+
filtered_text.append(word)
|
61 |
+
# Return the filtered text as a string
|
62 |
+
return ' '.join(filtered_text)
|
63 |
+
|
64 |
+
def generate_response(model, tokenizer, input_text, max_length=300, min_length=20, num_return_sequences=1, temperature=0.2, top_k=50, top_p=0.9, num_beams=10, repetition_penalty=1.0):
|
65 |
+
input_ids = tokenizer.encode(input_text, return_tensors='pt')
|
66 |
+
|
67 |
+
# Set model to eval mode
|
68 |
+
model.eval()
|
69 |
+
|
70 |
+
# Generate responses using different decoding strategies
|
71 |
+
try:
|
72 |
+
output = model.generate(
|
73 |
+
input_ids=input_ids,
|
74 |
+
max_length=max_length,
|
75 |
+
min_length=min_length,
|
76 |
+
num_return_sequences=num_return_sequences,
|
77 |
+
temperature=temperature,
|
78 |
+
top_k=top_k,
|
79 |
+
top_p=top_p,
|
80 |
+
num_beams=num_beams,
|
81 |
+
no_repeat_ngram_size=2,
|
82 |
+
do_sample=True,
|
83 |
+
repetition_penalty=repetition_penalty # Added repetition_penalty
|
84 |
+
)
|
85 |
+
|
86 |
+
# Decode the generated responses
|
87 |
+
responses = [tokenizer.decode(o, skip_special_tokens=True) for o in output]
|
88 |
+
# Remove input_text from the responses
|
89 |
+
responses = [response[len(tokenizer.decode(input_ids[0], skip_special_tokens=True)):] for response in responses]
|
90 |
+
|
91 |
+
# Filter out non-English words
|
92 |
+
responses = [filter_english_words(response) for response in responses]
|
93 |
+
|
94 |
+
return responses
|
95 |
+
|
96 |
+
except RuntimeError:
|
97 |
+
return ["I'm sorry, I encountered an error while generating a response."]
|
98 |
+
|
99 |
+
|
100 |
+
# Load pre-trained model and tokenizer
|
101 |
+
access_token = "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"
|
102 |
+
model_id = "Starcodium/VergilGPT2"
|
103 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id, revision="main", use_auth_token=access_token)
|
104 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, revision="main", use_auth_token=access_token)
|
105 |
+
|
106 |
+
tokenizer.pad_token = tokenizer.eos_token
|
107 |
+
model.config.pad_token_id = model.config.eos_token_id
|
108 |
+
|
109 |
+
# Get user input and generate responses
|
110 |
+
while True:
|
111 |
+
input_text = input("Type 'quit' or 'exit' to stop runtime.\nEnter your input text: ")
|
112 |
+
if input_text.lower() in ["quit", "exit"]:
|
113 |
+
break
|
114 |
+
|
115 |
+
responses = generate_response(model, tokenizer, input_text, max_length=100, min_length=20, num_return_sequences=1, temperature=0.7, top_k=40, top_p=0.5, num_beams=1, repetition_penalty=1.2)
|
116 |
+
responses = [r for r in responses if r.strip() != '']
|
117 |
+
if responses:
|
118 |
+
response = responses[0]
|
119 |
+
else:
|
120 |
+
response = "I'm sorry, I don't have a response at the moment."
|
121 |
+
|
122 |
+
# Print the bot's response
|
123 |
+
print("Vergil: "+response)
|
124 |
+
```
|
125 |
+
|
126 |
+
This code snippet allows you to engage in conversations with VergilGPT2. Simply enter your input text, and VergilGPT2 will generate responses based on the provided context. Experiment with different values of the variables temperature, top_k, and top_p to customize the response generation process according to your desired preferences.
|
127 |
+
|
128 |
+
Please note that the code assumes you have access to the Starcodium/VergilGPT2 model and its associated tokenizer. Ensure you have the required authentication token (access_token) to access the model and tokenizer.
|
129 |
+
|
130 |
## Installation
|
131 |
|
132 |
Make sure to install the required dependencies by running the following commands:
|