Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
5ca3297
1
Parent(s):
247d4bf
Formated code
Browse files
app.py
CHANGED
|
@@ -4,12 +4,28 @@ from transformers import Qwen2VLForConditionalGeneration, Qwen2VLProcessor
|
|
| 4 |
from qwen_vl_utils import process_vision_info
|
| 5 |
import torch
|
| 6 |
from PIL import Image
|
| 7 |
-
import subprocess
|
| 8 |
from datetime import datetime
|
| 9 |
import numpy as np
|
| 10 |
import os
|
| 11 |
|
| 12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
def array_to_image_path(image_array):
|
| 15 |
if image_array is None:
|
|
@@ -30,41 +46,9 @@ def array_to_image_path(image_array):
|
|
| 30 |
return full_path
|
| 31 |
|
| 32 |
|
| 33 |
-
model_id = "Qwen/Qwen2-VL-7B-Instruct"
|
| 34 |
-
model = Qwen2VLForConditionalGeneration.from_pretrained(
|
| 35 |
-
model_id,
|
| 36 |
-
device_map="auto",
|
| 37 |
-
torch_dtype=torch.bfloat16,
|
| 38 |
-
)
|
| 39 |
-
adapter_path = "sergiopaniego/qwen2-7b-instruct-trl-sft-ChartQA"
|
| 40 |
-
model.load_adapter(adapter_path)
|
| 41 |
-
|
| 42 |
-
processor = Qwen2VLProcessor.from_pretrained(model_id)
|
| 43 |
-
|
| 44 |
-
DESCRIPTION = """
|
| 45 |
-
# Qwen2-VL-7B-trl-sft-ChartQA Demo
|
| 46 |
-
|
| 47 |
-
This is a demo Space for a fine-tuned version of [Qwen2-VL-7B](https://huggingface.co/Qwen/Qwen2-VL-7B-Instruct) trained using [ChatQA dataset](https://huggingface.co/datasets/HuggingFaceM4/ChartQA).
|
| 48 |
-
|
| 49 |
-
The corresponding model is located [here](https://huggingface.co/sergiopaniego/qwen2-7b-instruct-trl-sft-ChartQA)
|
| 50 |
-
"""
|
| 51 |
-
|
| 52 |
-
kwargs = {}
|
| 53 |
-
kwargs['torch_dtype'] = torch.bfloat16
|
| 54 |
-
|
| 55 |
-
user_prompt = '<|user|>\n'
|
| 56 |
-
assistant_prompt = '<|assistant|>\n'
|
| 57 |
-
prompt_suffix = "<|end|>\n"
|
| 58 |
-
|
| 59 |
@spaces.GPU
|
| 60 |
def run_example(image, text_input=None):
|
| 61 |
image_path = array_to_image_path(image)
|
| 62 |
-
|
| 63 |
-
print(image_path)
|
| 64 |
-
#model = models[model_id]
|
| 65 |
-
#processor = processors[model_id]
|
| 66 |
-
|
| 67 |
-
prompt = f"{user_prompt}<|image_1|>\n{text_input}{prompt_suffix}{assistant_prompt}"
|
| 68 |
image = Image.fromarray(image).convert("RGB")
|
| 69 |
messages = [
|
| 70 |
{
|
|
@@ -121,13 +105,11 @@ with gr.Blocks(css=css) as demo:
|
|
| 121 |
with gr.Row():
|
| 122 |
with gr.Column():
|
| 123 |
input_img = gr.Image(label="Input Picture")
|
| 124 |
-
#model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value="sergiopaniego/qwen2-7b-instruct-trl-sft-ChartQA")
|
| 125 |
text_input = gr.Textbox(label="Question")
|
| 126 |
submit_btn = gr.Button(value="Submit")
|
| 127 |
with gr.Column():
|
| 128 |
output_text = gr.Textbox(label="Output Text")
|
| 129 |
|
| 130 |
-
#submit_btn.click(run_example, [input_img, text_input, model_selector], [output_text])
|
| 131 |
submit_btn.click(run_example, [input_img, text_input], [output_text])
|
| 132 |
|
| 133 |
demo.queue(api_open=False)
|
|
|
|
| 4 |
from qwen_vl_utils import process_vision_info
|
| 5 |
import torch
|
| 6 |
from PIL import Image
|
|
|
|
| 7 |
from datetime import datetime
|
| 8 |
import numpy as np
|
| 9 |
import os
|
| 10 |
|
| 11 |
|
| 12 |
+
DESCRIPTION = """
|
| 13 |
+
# Qwen2-VL-7B-trl-sft-ChartQA Demo
|
| 14 |
+
|
| 15 |
+
This is a demo Space for a fine-tuned version of [Qwen2-VL-7B](https://huggingface.co/Qwen/Qwen2-VL-7B-Instruct) trained using [ChatQA dataset](https://huggingface.co/datasets/HuggingFaceM4/ChartQA).
|
| 16 |
+
|
| 17 |
+
The corresponding model is located [here](https://huggingface.co/sergiopaniego/qwen2-7b-instruct-trl-sft-ChartQA).
|
| 18 |
+
"""
|
| 19 |
+
|
| 20 |
+
model_id = "Qwen/Qwen2-VL-7B-Instruct"
|
| 21 |
+
model = Qwen2VLForConditionalGeneration.from_pretrained(
|
| 22 |
+
model_id,
|
| 23 |
+
device_map="auto",
|
| 24 |
+
torch_dtype=torch.bfloat16,
|
| 25 |
+
)
|
| 26 |
+
adapter_path = "sergiopaniego/qwen2-7b-instruct-trl-sft-ChartQA"
|
| 27 |
+
model.load_adapter(adapter_path)
|
| 28 |
+
processor = Qwen2VLProcessor.from_pretrained(model_id)
|
| 29 |
|
| 30 |
def array_to_image_path(image_array):
|
| 31 |
if image_array is None:
|
|
|
|
| 46 |
return full_path
|
| 47 |
|
| 48 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
@spaces.GPU
|
| 50 |
def run_example(image, text_input=None):
|
| 51 |
image_path = array_to_image_path(image)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
image = Image.fromarray(image).convert("RGB")
|
| 53 |
messages = [
|
| 54 |
{
|
|
|
|
| 105 |
with gr.Row():
|
| 106 |
with gr.Column():
|
| 107 |
input_img = gr.Image(label="Input Picture")
|
|
|
|
| 108 |
text_input = gr.Textbox(label="Question")
|
| 109 |
submit_btn = gr.Button(value="Submit")
|
| 110 |
with gr.Column():
|
| 111 |
output_text = gr.Textbox(label="Output Text")
|
| 112 |
|
|
|
|
| 113 |
submit_btn.click(run_example, [input_img, text_input], [output_text])
|
| 114 |
|
| 115 |
demo.queue(api_open=False)
|