File size: 14,910 Bytes
2a23821 36916ef 2a23821 fdccfcb 2a23821 36916ef 2a23821 36916ef 2a23821 36916ef 2a23821 8cf3950 2a23821 7bc7781 2a23821 36916ef 2a23821 36916ef |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 |
---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen2-VL-2B-Instruct
pipeline_tag: image-text-to-text
library_name: transformers
tags:
- text-generation-inference
- Chem-Qwen
- chemistry
- InChI
- International Chemical Identifier
- IUPAC
---
# **ChemQwen2-vL [ Qwen for Chemistry Vision ]**
*ChemQwen2-vL* is a vision-language model fine-tuned based on the Qwen2VL-2B Instruct model. It has been trained using the International Chemical Identifier (InChI) format and is optimized for chemical component identification. The model excels at generating the InChI and providing descriptions of chemical compounds based on their images. Its architecture, derived from Qwen2-VL, can handle images of arbitrary resolutions by mapping them into a dynamic number of visual tokens. This offers a more human-like visual processing experience. The model operates in a multi-modal framework, combining image-text-text capabilities.
# **Key Enhancements**
* **State-of-the-art understanding of chemical images**: ChemQwen2-vL delivers top-tier performance in analyzing chemical images of varying resolutions and ratios, ensuring precise InChI generation and structural descriptions.
* **Advanced processing of chemical sequences**: Capable of handling complex molecular structures and long sequences, ChemQwen2-vL is ideal for tasks requiring detailed chemical reasoning and compound analysis.
* **Integration with automated laboratory systems**: With its powerful reasoning and decision-making capabilities, ChemQwen2-vL can integrate with laboratory robotics, enabling autonomous chemical analysis and reporting based on image and text inputs.
* **Multilingual Chemical Text Support**: To support global scientific communities, ChemQwen2-vL understands chemical texts in multiple languages, including English, Chinese, German, French, Japanese, Korean, Arabic, and more, for enhanced accessibility and collaboration.
# **Demo Inference With Documentation**
📌 *ChemQwen Demo :-* https://huggingface.co/prithivMLmods/ChemQwen2-vL/blob/main/Chemical%20Identifier%20Demo%20Documentation/Chemical%20Identifier%20Demo%20Documentation.ipynb
# **Datasets**
| Dataset Name | Description | InChI Link |
|--------------|-------------|------------|
| IUPAC Projects | A collection of datasets related to IUPAC projects, including chemical identifiers and standards. | [InChI Link](https://iupac.org/projects/) |
# **Use it with Transformers**
*Before using, ensure that the required libraries are successfully installed in the environment.*
!pip install gradio spaces transformers accelerate numpy requests torch torchvision qwen-vl-utils av ipython reportlab fpdf python-docx pillow huggingface_hub
*ChemQwen With Inference Documentation, **Before using, make sure that the `hf_token` is provided in the login field in the code below.***
```python
# Authenticate with Hugging Face
from huggingface_hub import login
# Log in to Hugging Face using the provided token
hf_token = '----xxxxx----'
login(hf_token)
# Demo
import gradio as gr
import spaces
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor, TextIteratorStreamer
from qwen_vl_utils import process_vision_info
import torch
from PIL import Image
import os
import uuid
import io
from threading import Thread
from reportlab.lib.pagesizes import A4
from reportlab.lib.styles import getSampleStyleSheet
from reportlab.lib import colors
from reportlab.platypus import SimpleDocTemplate, Image as RLImage, Paragraph, Spacer
from reportlab.pdfbase import pdfmetrics
from reportlab.pdfbase.ttfonts import TTFont
import docx
from docx.enum.text import WD_ALIGN_PARAGRAPH
# Define model options
MODEL_OPTIONS = {
"ChemQwen": "prithivMLmods/ChemQwen2-vL",
}
# Preload models and processors into CUDA
models = {}
processors = {}
for name, model_id in MODEL_OPTIONS.items():
print(f"Loading {name}...")
models[name] = Qwen2VLForConditionalGeneration.from_pretrained(
model_id,
trust_remote_code=True,
torch_dtype=torch.float16
).to("cuda").eval()
processors[name] = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
image_extensions = Image.registered_extensions()
def identify_and_save_blob(blob_path):
"""Identifies if the blob is an image and saves it."""
try:
with open(blob_path, 'rb') as file:
blob_content = file.read()
try:
Image.open(io.BytesIO(blob_content)).verify() # Check if it's a valid image
extension = ".png" # Default to PNG for saving
media_type = "image"
except (IOError, SyntaxError):
raise ValueError("Unsupported media type. Please upload a valid image.")
filename = f"temp_{uuid.uuid4()}_media{extension}"
with open(filename, "wb") as f:
f.write(blob_content)
return filename, media_type
except FileNotFoundError:
raise ValueError(f"The file {blob_path} was not found.")
except Exception as e:
raise ValueError(f"An error occurred while processing the file: {e}")
@spaces.GPU
def qwen_inference(model_name, media_input, text_input=None):
"""Handles inference for the selected model."""
model = models[model_name]
processor = processors[model_name]
if isinstance(media_input, str):
media_path = media_input
if media_path.endswith(tuple([i for i in image_extensions.keys()])):
media_type = "image"
else:
try:
media_path, media_type = identify_and_save_blob(media_input)
except Exception as e:
raise ValueError("Unsupported media type. Please upload a valid image.")
messages = [
{
"role": "user",
"content": [
{
"type": media_type,
media_type: media_path
},
{"type": "text", "text": text_input},
],
}
]
text = processor.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
image_inputs, _ = process_vision_info(messages)
inputs = processor(
text=[text],
images=image_inputs,
padding=True,
return_tensors="pt",
).to("cuda")
streamer = TextIteratorStreamer(
processor.tokenizer, skip_prompt=True, skip_special_tokens=True
)
generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)
thread = Thread(target=model.generate, kwargs=generation_kwargs)
thread.start()
buffer = ""
for new_text in streamer:
buffer += new_text
# Remove <|im_end|> or similar tokens from the output
buffer = buffer.replace("<|im_end|>", "")
yield buffer
def format_plain_text(output_text):
"""Formats the output text as plain text without LaTeX delimiters."""
# Remove LaTeX delimiters and convert to plain text
plain_text = output_text.replace("\\(", "").replace("\\)", "").replace("\\[", "").replace("\\]", "")
return plain_text
def generate_document(media_path, output_text, file_format, font_size, line_spacing, alignment, image_size):
"""Generates a document with the input image and plain text output."""
plain_text = format_plain_text(output_text)
if file_format == "pdf":
return generate_pdf(media_path, plain_text, font_size, line_spacing, alignment, image_size)
elif file_format == "docx":
return generate_docx(media_path, plain_text, font_size, line_spacing, alignment, image_size)
def generate_pdf(media_path, plain_text, font_size, line_spacing, alignment, image_size):
"""Generates a PDF document."""
filename = f"output_{uuid.uuid4()}.pdf"
doc = SimpleDocTemplate(
filename,
pagesize=A4,
rightMargin=inch,
leftMargin=inch,
topMargin=inch,
bottomMargin=inch
)
styles = getSampleStyleSheet()
styles["Normal"].fontSize = int(font_size)
styles["Normal"].leading = int(font_size) * line_spacing
styles["Normal"].alignment = {
"Left": 0,
"Center": 1,
"Right": 2,
"Justified": 4
}[alignment]
story = []
# Add image with size adjustment
image_sizes = {
"Small": (200, 200),
"Medium": (400, 400),
"Large": (600, 600)
}
img = RLImage(media_path, width=image_sizes[image_size][0], height=image_sizes[image_size][1])
story.append(img)
story.append(Spacer(1, 12))
# Add plain text output
text = Paragraph(plain_text, styles["Normal"])
story.append(text)
doc.build(story)
return filename
def generate_docx(media_path, plain_text, font_size, line_spacing, alignment, image_size):
"""Generates a DOCX document."""
filename = f"output_{uuid.uuid4()}.docx"
doc = docx.Document()
# Add image with size adjustment
image_sizes = {
"Small": docx.shared.Inches(2),
"Medium": docx.shared.Inches(4),
"Large": docx.shared.Inches(6)
}
doc.add_picture(media_path, width=image_sizes[image_size])
doc.add_paragraph()
# Add plain text output
paragraph = doc.add_paragraph()
paragraph.paragraph_format.line_spacing = line_spacing
paragraph.paragraph_format.alignment = {
"Left": WD_ALIGN_PARAGRAPH.LEFT,
"Center": WD_ALIGN_PARAGRAPH.CENTER,
"Right": WD_ALIGN_PARAGRAPH.RIGHT,
"Justified": WD_ALIGN_PARAGRAPH.JUSTIFY
}[alignment]
run = paragraph.add_run(plain_text)
run.font.size = docx.shared.Pt(int(font_size))
doc.save(filename)
return filename
# CSS for output styling
css = """
#output {
height: 500px;
overflow: auto;
border: 1px solid #ccc;
}
.submit-btn {
background-color: #cf3434 !important;
color: white !important;
}
.submit-btn:hover {
background-color: #ff2323 !important;
}
.download-btn {
background-color: #35a6d6 !important;
color: white !important;
}
.download-btn:hover {
background-color: #22bcff !important;
}
"""
# Gradio app setup
with gr.Blocks(css=css) as demo:
gr.Markdown("# ChemQwen Chemical Identifier")
with gr.Tab(label="Image Input"):
with gr.Row():
with gr.Column():
model_choice = gr.Dropdown(
label="Model Selection",
choices=list(MODEL_OPTIONS.keys()),
value="ChemQwen"
)
input_media = gr.File(
label="Upload Image", type="filepath"
)
text_input = gr.Textbox(label="Question", placeholder="Ask a question about the image...")
submit_btn = gr.Button(value="Submit", elem_classes="submit-btn")
with gr.Column():
output_text = gr.Textbox(label="Output Text", lines=10)
plain_text_output = gr.Textbox(label="Standardized Plain Text", lines=10)
submit_btn.click(
qwen_inference, [model_choice, input_media, text_input], [output_text]
).then(
lambda output_text: format_plain_text(output_text), [output_text], [plain_text_output]
)
# Add examples directly usable by clicking
with gr.Row():
with gr.Column():
line_spacing = gr.Dropdown(
choices=[0.5, 1.0, 1.15, 1.5, 2.0, 2.5, 3.0],
value=1.5,
label="Line Spacing"
)
font_size = gr.Dropdown(
choices=["8", "10", "12", "14", "16", "18", "20", "22", "24"],
value="18",
label="Font Size"
)
alignment = gr.Dropdown(
choices=["Left", "Center", "Right", "Justified"],
value="Justified",
label="Text Alignment"
)
image_size = gr.Dropdown(
choices=["Small", "Medium", "Large"],
value="Small",
label="Image Size"
)
file_format = gr.Radio(["pdf", "docx"], label="File Format", value="pdf")
get_document_btn = gr.Button(value="Get Document", elem_classes="download-btn")
get_document_btn.click(
generate_document, [input_media, output_text, file_format, font_size, line_spacing, alignment, image_size], gr.File(label="Download Document")
)
demo.launch(debug=True)
```
# **Intended Use**
ChemQwen2-vL is designed for applications in the field of cheminformatics and chemical research:
1. **Chemical Structure Recognition**:
- Generating precise International Chemical Identifiers (InChI) from chemical images.
- Providing detailed descriptions of molecular structures based on visual inputs.
2. **Scientific Research and Education**:
- Supporting researchers and students by offering automated analysis of chemical diagrams and visual materials.
3. **Integration with Laboratory Systems**:
- Assisting in automation by integrating with robotic systems for chemical analysis, reporting, and data generation.
4. **Multilingual Chemical Data Analysis**:
- Supporting international collaborations with multilingual recognition of chemical texts and images.
5. **Advanced Question Answering**:
- Enabling chemistry-related visual question answering and reasoning for educational tools, publications, and research platforms.
# **Limitations**
1. **Accuracy for Uncommon or Complex Structures**:
- May struggle with highly complex or unconventional chemical diagrams not adhering to standard formats.
2. **Dependency on Input Quality**:
- Performance can degrade with low-resolution, poorly scanned, or incomplete images.
3. **Limited Domain Knowledge**:
- While proficient in chemical component identification, the model might lack domain-specific context in specialized subfields of chemistry.
4. **Processing Speed for Large Data Sets**:
- Performance may slow down when processing very large datasets or videos of extended duration.
5. **Hardware Requirements**:
- Requires high computational resources, including GPUs, for optimal performance during inference.
6. **Language and Symbol Constraints**:
- Though multilingual, rare or specialized symbols and annotations may not always be recognized accurately.
7. **No Real-Time Laboratory Decisions**:
- While it integrates with automated systems, real-time critical decisions still require human validation. |