Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
import gradio as gr
|
3 |
+
import torch
|
4 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
5 |
+
from peft import PeftModel
|
6 |
+
|
7 |
+
from functools import lru_cache
|
8 |
+
|
9 |
+
# Define models
|
10 |
+
BASE_MODEL = "deepseek-ai/deepseek-math-7b-rl"
|
11 |
+
FINETUNED_MODEL = "LaibaIrfan/emoji_math"
|
12 |
+
|
13 |
+
# Load tokenizer and model
|
14 |
+
@lru_cache()
|
15 |
+
def load_model():
|
16 |
+
tokenizer = AutoTokenizer.from_pretrained(FINETUNED_MODEL)
|
17 |
+
|
18 |
+
base_model = AutoModelForCausalLM.from_pretrained(
|
19 |
+
BASE_MODEL,
|
20 |
+
torch_dtype=torch.float16, # Use float16 for efficiency
|
21 |
+
device_map="auto", # Auto-assign device (GPU if available)
|
22 |
+
load_in_8bit=True # Reduce memory usage (slightly increases inference time)
|
23 |
+
)
|
24 |
+
|
25 |
+
model = PeftModel.from_pretrained(
|
26 |
+
base_model,
|
27 |
+
FINETUNED_MODEL,
|
28 |
+
device_map="auto"
|
29 |
+
)
|
30 |
+
|
31 |
+
return tokenizer, model
|
32 |
+
|
33 |
+
# Load the model
|
34 |
+
tokenizer, model = load_model()
|
35 |
+
|
36 |
+
# Function to generate the result
|
37 |
+
def generate_result(incorrect_math):
|
38 |
+
input_text = f"Incorrect: {incorrect_math}\nCorrect:"
|
39 |
+
|
40 |
+
# Move input to GPU
|
41 |
+
inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
|
42 |
+
|
43 |
+
# Generate output on GPU
|
44 |
+
output = model.generate(**inputs, max_length=200)
|
45 |
+
|
46 |
+
return tokenizer.decode(output[0], skip_special_tokens=True)
|
47 |
+
|
48 |
+
# Gradio Interface
|
49 |
+
iface = gr.Interface(
|
50 |
+
fn=generate_result,
|
51 |
+
inputs="text",
|
52 |
+
outputs="text",
|
53 |
+
title="Emoji Math Solver 🧮",
|
54 |
+
description="Enter an emoji-based math equation, and the model will generate the correct answer!"
|
55 |
+
)
|
56 |
+
|
57 |
+
iface.launch(debug=True, share=True, inline=True)
|
58 |
+
|
59 |
+
# Function to generate result
|
60 |
+
def generate_result(incorrect_math):
|
61 |
+
input_text = f"Incorrect: {incorrect_math}\nCorrect:"
|
62 |
+
inputs = tokenizer(input_text, return_tensors="pt").to("cuda") # Use GPU if available
|
63 |
+
output = model.generate(**inputs, max_length=200)
|
64 |
+
return tokenizer.decode(output[0], skip_special_tokens=True)
|
65 |
+
|
66 |
+
# Gradio Interface
|
67 |
+
iface = gr.Interface(
|
68 |
+
fn=generate_result,
|
69 |
+
inputs="text",
|
70 |
+
outputs="text",
|
71 |
+
title="Emoji Math Solver 🧮",
|
72 |
+
description="Enter an emoji-based math equation, and the model will generate the correct answer!"
|
73 |
+
)
|
74 |
+
|
75 |
+
iface.launch(share=True)
|