Spaces:
Runtime error
Runtime error
Ravinandan
commited on
Commit
•
dcfbbe4
1
Parent(s):
0823dd5
Update app.py
Browse files
app.py
CHANGED
@@ -1,11 +1,8 @@
|
|
1 |
import torch
|
2 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
3 |
import gradio as gr
|
4 |
|
5 |
-
# Set environment variable for PyTorch CUDA memory management
|
6 |
-
import os
|
7 |
-
os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'expandable_segments:True'
|
8 |
-
|
9 |
# Load the model and tokenizer
|
10 |
model = AutoModelForCausalLM.from_pretrained(
|
11 |
"qresearch/llama-3.1-8B-vision-378",
|
@@ -17,21 +14,9 @@ tokenizer = AutoTokenizer.from_pretrained("qresearch/llama-3.1-8B-vision-378", u
|
|
17 |
|
18 |
# Define the function to process the image and instruction
|
19 |
def describe_image(image, instruction):
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
# Generate a description using the model
|
24 |
-
with torch.no_grad(): # Avoid storing gradients to save memory
|
25 |
-
outputs = model.generate(
|
26 |
-
**inputs,
|
27 |
-
max_new_tokens=128,
|
28 |
-
do_sample=True,
|
29 |
-
temperature=0.3
|
30 |
-
)
|
31 |
-
|
32 |
-
# Decode the generated tokens to a string
|
33 |
-
description = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
34 |
-
|
35 |
return description
|
36 |
|
37 |
# Create the Gradio interface
|
@@ -42,9 +27,9 @@ interface = gr.Interface(
|
|
42 |
gr.Textbox(placeholder="Enter your instruction here...", label="Instruction") # Input for the instruction
|
43 |
],
|
44 |
outputs="text", # Output is text (the description)
|
45 |
-
title="LLaMA 3.1 with
|
46 |
description="Upload an image and enter an instruction to generate a description based on the provided instruction."
|
47 |
)
|
48 |
|
49 |
# Launch the Gradio app
|
50 |
-
interface.launch()
|
|
|
1 |
import torch
|
2 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
3 |
+
from PIL import Image
|
4 |
import gradio as gr
|
5 |
|
|
|
|
|
|
|
|
|
6 |
# Load the model and tokenizer
|
7 |
model = AutoModelForCausalLM.from_pretrained(
|
8 |
"qresearch/llama-3.1-8B-vision-378",
|
|
|
14 |
|
15 |
# Define the function to process the image and instruction
|
16 |
def describe_image(image, instruction):
|
17 |
+
description = model.answer_question(
|
18 |
+
image, instruction, tokenizer, max_new_tokens=128, do_sample=True, temperature=0.3
|
19 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
return description
|
21 |
|
22 |
# Create the Gradio interface
|
|
|
27 |
gr.Textbox(placeholder="Enter your instruction here...", label="Instruction") # Input for the instruction
|
28 |
],
|
29 |
outputs="text", # Output is text (the description)
|
30 |
+
title="LLaMA 3.1 with vision",
|
31 |
description="Upload an image and enter an instruction to generate a description based on the provided instruction."
|
32 |
)
|
33 |
|
34 |
# Launch the Gradio app
|
35 |
+
interface.launch()
|