Update README.md
Browse files
README.md
CHANGED
@@ -9,4 +9,106 @@ base_model:
|
|
9 |
- Qwen/Qwen2.5-VL-3B-Instruct
|
10 |
pipeline_tag: image-text-to-text
|
11 |
library_name: transformers
|
12 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
- Qwen/Qwen2.5-VL-3B-Instruct
|
10 |
pipeline_tag: image-text-to-text
|
11 |
library_name: transformers
|
12 |
+
---
|
13 |
+
|
14 |
+

|
15 |
+
|
16 |
+
# **Behemoth-3B-070225-post0.1**
|
17 |
+
|
18 |
+
> The **Behemoth-3B-070225-post0.1** model is a fine-tuned version of **Qwen2.5-VL-3B-Instruct**, optimized for **Detailed Image Captioning**, **OCR Tasks**, and **Chain-of-Thought Reasoning**. Built on top of the Qwen2.5-VL architecture, this model enhances visual understanding capabilities with focused training on the 50k LLaVA-CoT-o1-Instruct dataset for superior image analysis and detailed reasoning tasks.
|
19 |
+
|
20 |
+
# Key Enhancements
|
21 |
+
|
22 |
+
* **Detailed Image Captioning**: Advanced capability for generating comprehensive, contextually rich descriptions of visual content with fine-grained detail recognition.
|
23 |
+
|
24 |
+
* **Enhanced OCR Performance**: Designed to efficiently extract and recognize text from images with high accuracy across various fonts, layouts, and image qualities.
|
25 |
+
|
26 |
+
* **Chain-of-Thought Reasoning**: Specialized in providing step-by-step logical reasoning processes for complex visual analysis tasks, breaking down problems into manageable components.
|
27 |
+
|
28 |
+
* **Superior Visual Understanding**: Optimized for precise interpretation of visual elements, spatial relationships, and contextual information within images.
|
29 |
+
|
30 |
+
* **Instruction Following**: Enhanced ability to follow detailed instructions for specific image analysis tasks while maintaining reasoning transparency.
|
31 |
+
|
32 |
+
* **State-of-the-Art Performance on Vision Tasks**: Achieves competitive results on visual question answering, image captioning, and OCR benchmarks.
|
33 |
+
|
34 |
+
* **Efficient 3B Parameter Model**: Provides strong performance while maintaining computational efficiency for broader accessibility.
|
35 |
+
|
36 |
+
* **Multi-Modal Reasoning**: Enables comprehensive analysis combining visual perception with logical reasoning chains.
|
37 |
+
|
38 |
+
# Quick Start with Transformers
|
39 |
+
|
40 |
+
```python
|
41 |
+
from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
|
42 |
+
from qwen_vl_utils import process_vision_info
|
43 |
+
|
44 |
+
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
45 |
+
"prithivMLmods/Behemoth-3B-070225-post0.1", torch_dtype="auto", device_map="auto"
|
46 |
+
)
|
47 |
+
|
48 |
+
processor = AutoProcessor.from_pretrained("prithivMLmods/Behemoth-3B-070225-post0.1")
|
49 |
+
|
50 |
+
messages = [
|
51 |
+
{
|
52 |
+
"role": "user",
|
53 |
+
"content": [
|
54 |
+
{
|
55 |
+
"type": "image",
|
56 |
+
"image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
|
57 |
+
},
|
58 |
+
{"type": "text", "text": "Provide a detailed caption for this image and explain your reasoning step by step."},
|
59 |
+
],
|
60 |
+
}
|
61 |
+
]
|
62 |
+
|
63 |
+
text = processor.apply_chat_template(
|
64 |
+
messages, tokenize=False, add_generation_prompt=True
|
65 |
+
)
|
66 |
+
image_inputs, video_inputs = process_vision_info(messages)
|
67 |
+
inputs = processor(
|
68 |
+
text=[text],
|
69 |
+
images=image_inputs,
|
70 |
+
videos=video_inputs,
|
71 |
+
padding=True,
|
72 |
+
return_tensors="pt",
|
73 |
+
)
|
74 |
+
inputs = inputs.to("cuda")
|
75 |
+
|
76 |
+
generated_ids = model.generate(**inputs, max_new_tokens=256)
|
77 |
+
generated_ids_trimmed = [
|
78 |
+
out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
79 |
+
]
|
80 |
+
output_text = processor.batch_decode(
|
81 |
+
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
|
82 |
+
)
|
83 |
+
print(output_text)
|
84 |
+
```
|
85 |
+
|
86 |
+
# Intended Use
|
87 |
+
|
88 |
+
This model is intended for:
|
89 |
+
|
90 |
+
* **Detailed Image Captioning**: Generating comprehensive, nuanced descriptions of visual content for accessibility, content creation, and analysis purposes.
|
91 |
+
* **OCR Applications**: High-accuracy text extraction from images, documents, signs, and handwritten content.
|
92 |
+
* **Chain-of-Thought Visual Analysis**: Providing step-by-step reasoning for complex visual interpretation tasks.
|
93 |
+
* **Educational Content Creation**: Generating detailed explanations of visual materials with logical reasoning chains.
|
94 |
+
* **Content Accessibility**: Creating detailed alt-text and descriptions for visually impaired users.
|
95 |
+
* **Visual Question Answering**: Answering complex questions about images with detailed reasoning processes.
|
96 |
+
* **Document Analysis**: Processing and understanding visual documents with both text extraction and content comprehension.
|
97 |
+
* **Research and Analysis**: Supporting academic and professional research requiring detailed visual analysis with transparent reasoning.
|
98 |
+
|
99 |
+
# Base Training Details
|
100 |
+
|
101 |
+
* **Base Model**: Qwen2.5-VL-3B-Instruct
|
102 |
+
* **Training Dataset**: 50k LLaVA-CoT-o1-Instruct dataset
|
103 |
+
* **Specialized Training Focus**: Chain-of-thought reasoning, detailed captioning, and OCR tasks
|
104 |
+
* **Model Size**: 3 billion parameters for efficient deployment
|
105 |
+
|
106 |
+
# Limitations
|
107 |
+
|
108 |
+
* **Computational Requirements**: While more efficient than larger models, still requires adequate GPU memory for optimal performance.
|
109 |
+
* **Image Quality Sensitivity**: Performance may degrade on extremely low-quality, heavily occluded, or severely distorted images.
|
110 |
+
* **Processing Speed**: Chain-of-thought reasoning may result in longer response times compared to direct answer models.
|
111 |
+
* **Language Coverage**: Primarily optimized for English language tasks, with variable performance on other languages.
|
112 |
+
* **Context Length**: Limited by the base model's context window for very long reasoning chains.
|
113 |
+
* **Hallucination Risk**: May occasionally generate plausible but incorrect details, especially in ambiguous visual scenarios.
|
114 |
+
* **Resource Constraints**: Not optimized for real-time applications on edge devices or low-resource environments.
|