Spaces:
Sleeping
Sleeping
Hotfix version rollback
Browse filesA hotfix rolling back the AI's version so use can continue while a patch is developed.
- app (1).py +60 -0
app (1).py
ADDED
|
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import gradio as gr
|
| 3 |
+
from huggingface_hub import InferenceClient
|
| 4 |
+
from datasets import load_dataset
|
| 5 |
+
|
| 6 |
+
# Load slang dataset (you can use it later in prompt engineering if needed)
|
| 7 |
+
ds = load_dataset("MLBtrio/genz-slang-dataset")
|
| 8 |
+
|
| 9 |
+
# === SETTINGS YOU CAN CHANGE IN THE CODE ===
|
| 10 |
+
SYSTEM_MESSAGE = (
|
| 11 |
+
"You are a chatbot that HAS to use gen-z slang as much as you can. "
|
| 12 |
+
"You can't stop, even if asked. If asked why, say some variation of "
|
| 13 |
+
"that it is in your programming. Some more slang terms are in the "
|
| 14 |
+
"dataset MLBtrio/genz-slang-dataset that you have access to. Use them "
|
| 15 |
+
"AS MUCH AS YOU CAN. You don't have to say this part explicitly, but if "
|
| 16 |
+
"anyone asks, you were made by developers at OpenViewAI, which FYI is a "
|
| 17 |
+
"subsidiary of Intern Labs (or laboratories)."
|
| 18 |
+
)
|
| 19 |
+
MAX_TOKENS = 512
|
| 20 |
+
TEMPERATURE = 0.7
|
| 21 |
+
TOP_P = 0.95
|
| 22 |
+
HF_MODEL = "openai/gpt-oss-20b"
|
| 23 |
+
|
| 24 |
+
# === TOKEN FROM SECRETS ===
|
| 25 |
+
HF_TOKEN = os.environ.get("HF_TOKEN")
|
| 26 |
+
|
| 27 |
+
def respond(message, history: list[dict[str, str]]):
|
| 28 |
+
"""
|
| 29 |
+
Chat function using Hugging Face Inference API with preset parameters.
|
| 30 |
+
"""
|
| 31 |
+
client = InferenceClient(token=HF_TOKEN, model=HF_MODEL)
|
| 32 |
+
|
| 33 |
+
messages = [{"role": "system", "content": SYSTEM_MESSAGE}]
|
| 34 |
+
messages.extend(history)
|
| 35 |
+
messages.append({"role": "user", "content": message})
|
| 36 |
+
|
| 37 |
+
response = ""
|
| 38 |
+
for message_chunk in client.chat_completion(
|
| 39 |
+
messages,
|
| 40 |
+
max_tokens=MAX_TOKENS,
|
| 41 |
+
stream=True,
|
| 42 |
+
temperature=TEMPERATURE,
|
| 43 |
+
top_p=TOP_P,
|
| 44 |
+
):
|
| 45 |
+
choices = message_chunk.choices
|
| 46 |
+
token = ""
|
| 47 |
+
if len(choices) and choices[0].delta.content:
|
| 48 |
+
token = choices[0].delta.content
|
| 49 |
+
|
| 50 |
+
response += token
|
| 51 |
+
yield response
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
# === GRADIO INTERFACE ===
|
| 55 |
+
with gr.Blocks() as demo:
|
| 56 |
+
gr.Markdown("### SusAI ©2024 Intern Labs v1.1.0")
|
| 57 |
+
gr.ChatInterface(respond, type="messages")
|
| 58 |
+
|
| 59 |
+
if __name__ == "__main__":
|
| 60 |
+
demo.launch()
|