Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
4 |
+
|
5 |
+
# MODEL REPO
|
6 |
+
MODEL_NAME = "mistralai/Mistral-7B-v0.1"
|
7 |
+
|
8 |
+
# Load tokenizer
|
9 |
+
print("Loading tokenizer...")
|
10 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
11 |
+
MODEL_NAME,
|
12 |
+
trust_remote_code=True
|
13 |
+
)
|
14 |
+
|
15 |
+
# Load model in 4-bit on CPU
|
16 |
+
# (Even though we set device_map="auto", on a free Space there's no GPU, so it stays on CPU.)
|
17 |
+
print("Loading model in 4-bit...")
|
18 |
+
model = AutoModelForCausalLM.from_pretrained(
|
19 |
+
MODEL_NAME,
|
20 |
+
torch_dtype=torch.float16,
|
21 |
+
device_map="auto", # auto-detect available devices
|
22 |
+
load_in_4bit=True, # bitsandbytes for quantization
|
23 |
+
trust_remote_code=True # Mistral uses custom code
|
24 |
+
)
|
25 |
+
|
26 |
+
model.eval()
|
27 |
+
|
28 |
+
def chat_mistral(prompt):
|
29 |
+
"""
|
30 |
+
Generates a response from Mistral 7B given a user prompt.
|
31 |
+
"""
|
32 |
+
# Tokenize
|
33 |
+
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
34 |
+
|
35 |
+
# Generate
|
36 |
+
with torch.no_grad():
|
37 |
+
outputs = model.generate(
|
38 |
+
**inputs,
|
39 |
+
max_new_tokens=128, # limit output length to avoid OOM
|
40 |
+
temperature=0.7,
|
41 |
+
repetition_penalty=1.1
|
42 |
+
)
|
43 |
+
|
44 |
+
# Decode
|
45 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
46 |
+
return response
|
47 |
+
|
48 |
+
# Create a Gradio interface
|
49 |
+
demo = gr.Interface(
|
50 |
+
fn=chat_mistral,
|
51 |
+
inputs=gr.Textbox(lines=3, label="Your Prompt"),
|
52 |
+
outputs=gr.Textbox(label="Mistral 7B Response"),
|
53 |
+
title="Mistral 7B (4-bit) Chat",
|
54 |
+
description=(
|
55 |
+
"A minimal Mistral-7B demo running on free CPU. "
|
56 |
+
"Inference will be slow and might run out of memory. "
|
57 |
+
"Use short prompts!"
|
58 |
+
)
|
59 |
+
)
|
60 |
+
|
61 |
+
# Launch the Gradio app
|
62 |
+
if __name__ == "__main__":
|
63 |
+
demo.launch()
|