Upload README.md
Browse files
README.md
CHANGED
@@ -1,199 +1,322 @@
|
|
1 |
---
|
|
|
|
|
|
|
2 |
library_name: transformers
|
3 |
-
tags:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
---
|
|
|
5 |
|
6 |
-
|
7 |
|
8 |
-
|
9 |
|
|
|
10 |
|
|
|
11 |
|
12 |
-
|
13 |
|
14 |
-
|
15 |
|
16 |
-
|
17 |
|
18 |
-
|
19 |
|
20 |
-
-
|
21 |
-
- **Funded by [optional]:** [More Information Needed]
|
22 |
-
- **Shared by [optional]:** [More Information Needed]
|
23 |
-
- **Model type:** [More Information Needed]
|
24 |
-
- **Language(s) (NLP):** [More Information Needed]
|
25 |
-
- **License:** [More Information Needed]
|
26 |
-
- **Finetuned from model [optional]:** [More Information Needed]
|
27 |
|
28 |
-
|
|
|
|
|
|
|
|
|
|
|
29 |
|
30 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
|
32 |
-
|
33 |
-
- **Paper [optional]:** [More Information Needed]
|
34 |
-
- **Demo [optional]:** [More Information Needed]
|
35 |
|
36 |
-
## Uses
|
37 |
|
38 |
-
|
39 |
|
40 |
-
|
41 |
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
|
55 |
-
|
56 |
-
[More Information Needed]
|
57 |
-
|
58 |
-
## Bias, Risks, and Limitations
|
59 |
-
|
60 |
-
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
61 |
-
|
62 |
-
[More Information Needed]
|
63 |
-
|
64 |
-
### Recommendations
|
65 |
-
|
66 |
-
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
|
67 |
-
|
68 |
-
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
|
69 |
-
|
70 |
-
## How to Get Started with the Model
|
71 |
-
|
72 |
-
Use the code below to get started with the model.
|
73 |
-
|
74 |
-
[More Information Needed]
|
75 |
-
|
76 |
-
## Training Details
|
77 |
-
|
78 |
-
### Training Data
|
79 |
-
|
80 |
-
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
81 |
-
|
82 |
-
[More Information Needed]
|
83 |
-
|
84 |
-
### Training Procedure
|
85 |
-
|
86 |
-
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
|
87 |
-
|
88 |
-
#### Preprocessing [optional]
|
89 |
-
|
90 |
-
[More Information Needed]
|
91 |
-
|
92 |
-
|
93 |
-
#### Training Hyperparameters
|
94 |
-
|
95 |
-
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
|
96 |
-
|
97 |
-
#### Speeds, Sizes, Times [optional]
|
98 |
-
|
99 |
-
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
100 |
-
|
101 |
-
[More Information Needed]
|
102 |
-
|
103 |
-
## Evaluation
|
104 |
-
|
105 |
-
<!-- This section describes the evaluation protocols and provides the results. -->
|
106 |
-
|
107 |
-
### Testing Data, Factors & Metrics
|
108 |
-
|
109 |
-
#### Testing Data
|
110 |
-
|
111 |
-
<!-- This should link to a Dataset Card if possible. -->
|
112 |
-
|
113 |
-
[More Information Needed]
|
114 |
-
|
115 |
-
#### Factors
|
116 |
-
|
117 |
-
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
|
118 |
-
|
119 |
-
[More Information Needed]
|
120 |
-
|
121 |
-
#### Metrics
|
122 |
-
|
123 |
-
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
124 |
-
|
125 |
-
[More Information Needed]
|
126 |
-
|
127 |
-
### Results
|
128 |
-
|
129 |
-
[More Information Needed]
|
130 |
-
|
131 |
-
#### Summary
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
## Model Examination [optional]
|
136 |
-
|
137 |
-
<!-- Relevant interpretability work for the model goes here -->
|
138 |
-
|
139 |
-
[More Information Needed]
|
140 |
-
|
141 |
-
## Environmental Impact
|
142 |
-
|
143 |
-
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
|
144 |
-
|
145 |
-
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
146 |
-
|
147 |
-
- **Hardware Type:** [More Information Needed]
|
148 |
-
- **Hours used:** [More Information Needed]
|
149 |
-
- **Cloud Provider:** [More Information Needed]
|
150 |
-
- **Compute Region:** [More Information Needed]
|
151 |
-
- **Carbon Emitted:** [More Information Needed]
|
152 |
-
|
153 |
-
## Technical Specifications [optional]
|
154 |
-
|
155 |
-
### Model Architecture and Objective
|
156 |
-
|
157 |
-
[More Information Needed]
|
158 |
-
|
159 |
-
### Compute Infrastructure
|
160 |
-
|
161 |
-
[More Information Needed]
|
162 |
-
|
163 |
-
#### Hardware
|
164 |
-
|
165 |
-
[More Information Needed]
|
166 |
-
|
167 |
-
#### Software
|
168 |
-
|
169 |
-
[More Information Needed]
|
170 |
-
|
171 |
-
## Citation [optional]
|
172 |
-
|
173 |
-
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
174 |
-
|
175 |
-
**BibTeX:**
|
176 |
-
|
177 |
-
[More Information Needed]
|
178 |
-
|
179 |
-
**APA:**
|
180 |
-
|
181 |
-
[More Information Needed]
|
182 |
-
|
183 |
-
## Glossary [optional]
|
184 |
-
|
185 |
-
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
186 |
-
|
187 |
-
[More Information Needed]
|
188 |
-
|
189 |
-
## More Information [optional]
|
190 |
-
|
191 |
-
[More Information Needed]
|
192 |
-
|
193 |
-
## Model Card Authors [optional]
|
194 |
-
|
195 |
-
[More Information Needed]
|
196 |
-
|
197 |
-
## Model Card Contact
|
198 |
-
|
199 |
-
[More Information Needed]
|
|
|
1 |
---
|
2 |
+
language:
|
3 |
+
- multilingual
|
4 |
+
license: apache-2.0
|
5 |
library_name: transformers
|
6 |
+
tags:
|
7 |
+
- nlp
|
8 |
+
- code
|
9 |
+
- vision
|
10 |
+
- chemistry
|
11 |
+
- engineering
|
12 |
+
- biology
|
13 |
+
- bio-inspired
|
14 |
+
- text-generation-inference
|
15 |
+
- materials science
|
16 |
+
pipeline_tag: image-text-to-text
|
17 |
+
inference:
|
18 |
+
parameters:
|
19 |
+
temperature: 0.3
|
20 |
+
widget:
|
21 |
+
- messages:
|
22 |
+
- role: user
|
23 |
+
content: <|image_1|>Can you describe what you see in the image?
|
24 |
---
|
25 |
+
## Model Summary
|
26 |
|
27 |
+
Cephalo is a series of multimodal materials science focused vision large language models (V-LLMs) designed to integrate visual and linguistic data for advanced understanding and interaction in human-AI or multi-agent AI frameworks.
|
28 |
|
29 |
+
A novel aspect of Cephalo's development is the innovative dataset generation method. The extraction process employs advanced algorithms to accurately detect and separate images and their corresponding textual descriptions from complex PDF documents. It involves extracting images and captions from PDFs to create well-reasoned image-text pairs, utilizing large language models (LLMs) for natural language processing. These image-text pairs are then refined and validated through LLM-based NLP processing, ensuring high-quality and contextually relevant data for training.
|
30 |
|
31 |
+
Cephalo can interpret complex visual scenes and generating contextually accurate language descriptions and answer queries.
|
32 |
|
33 |
+
The model is developed to process diverse inputs, including images and text, facilitating a broad range of applications such as image captioning, visual question answering, and multimodal content generation. The architecture combines a vision encoder model and an autoregressive transformer to process complex natural language understanding.
|
34 |
|
35 |
+
![image/png](https://cdn-uploads.huggingface.co/production/uploads/623ce1c6b66fedf374859fe7/kl5GWBP9WS0D4uwd1t3S7.png)
|
36 |
|
37 |
+
Cephalo provides a robust framework for multimodal interaction and understanding, including the development of complex generative pipelines to create 2D and 3D renderings of material microstructures as input for additive manufacturing methods.
|
38 |
|
39 |
+
This version of Cephalo, lamm-mit/Cephalo-Idefics-2-vision-8b-alpha, is based on the HuggingFaceM4/idefics2-8b-chatty model. The model was trained on a combination of scientific text-image data extracted from Wikipedia and scientific papers. For further details on the base model, see: https://huggingface.co/HuggingFaceM4/idefics2-8b-chatty. More details about technical aspects of the model, training and example applications to materials science problems are provided in the paper (reference at the bottom).
|
40 |
|
41 |
+
### Chat Format
|
42 |
|
43 |
+
The lamm-mit/Cephalo-Idefics-2-vision-8b-alpha is suiteable for one or more image inputs, wih prompts using the chat format as follows:
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
|
45 |
+
```raw
|
46 |
+
User: You carefully study the image, and respond accurately, but succinctly. Think step-by-step.
|
47 |
+
<image>What is shown in this image, and what is the relevance for materials design? Include a discussion of multi-agent AI.<end_of_utterance>
|
48 |
+
Assistant:
|
49 |
+
```
|
50 |
+
where the model generates the text after `Assistant:` . For multi-turn conversations, the prompt should be formatted as follows:
|
51 |
|
52 |
+
```raw
|
53 |
+
User: You carefully study the image, and respond accurately, but succinctly. Think step-by-step.
|
54 |
+
<image>What is shown in this image, and what is the relevance for materials design? Include a discussion of multi-agent AI.<end_of_utterance>
|
55 |
+
Assistant: The image depicts ants climbing a vertical surface using their legs and claws. This behavior is observed in nature and can inspire the design of multi-agent AI systems that mimic the coordinated movement of these insects. The relevance lies in the potential application of such systems in robotics and materials science, where efficient and adaptive movement is crucial.<end_of_utterance>
|
56 |
+
User: How could this be used to design a fracture resistant material?<end_of_utterance>
|
57 |
+
Assistant:
|
58 |
+
```
|
59 |
+
|
60 |
+
If you need to manually set the chat template:
|
61 |
+
|
62 |
+
```
|
63 |
+
IDEFICS2_CHAT_TEMPLATE = "{% for message in messages %}{{message['role'].capitalize()}}{% if message['content'][0]['type'] == 'image' %}{{':'}}{% else %}{{': '}}{% endif %}{% for line in message['content'] %}{% if line['type'] == 'text' %}{{line['text']}}{% elif line['type'] == 'image' %}{{ '<image>' }}{% endif %}{% endfor %}<end_of_utterance>\n{% endfor %}{% if add_generation_prompt %}{{ 'Assistant:' }}{% endif %}"
|
64 |
+
```
|
65 |
+
|
66 |
+
### Sample inference code
|
67 |
+
|
68 |
+
This code snippets show how to get quickly started on a GPU:
|
69 |
+
|
70 |
+
```python
|
71 |
+
from PIL import Image
|
72 |
+
import requests
|
73 |
+
|
74 |
+
DEVICE='cuda:0'
|
75 |
+
|
76 |
+
from transformers import AutoProcessor, Idefics2ForConditionalGeneration
|
77 |
+
from tqdm.notebook import tqdm
|
78 |
+
|
79 |
+
model_id='lamm-mit/Cephalo-Idefics-2-vision-8b-alpha'
|
80 |
+
|
81 |
+
model = Idefics2ForConditionalGeneration.from_pretrained( model_id,
|
82 |
+
torch_dtype=torch.bfloat16, #if your GPU allows
|
83 |
+
_attn_implementation="flash_attention_2", #make sure Flash Attention 2 is installed
|
84 |
+
trust_remote_code=True,
|
85 |
+
).to (DEVICE)
|
86 |
+
processor = AutoProcessor.from_pretrained(
|
87 |
+
f"{model_id}",
|
88 |
+
do_image_splitting=True
|
89 |
+
)
|
90 |
+
```
|
91 |
+
See section towards the end for more comments on model optimization, including quantization.
|
92 |
+
|
93 |
+
|
94 |
+
If you need to manually set the chat template:
|
95 |
+
|
96 |
+
```python
|
97 |
+
IDEFICS2_CHAT_TEMPLATE = "{% for message in messages %}{{message['role'].capitalize()}}{% if message['content'][0]['type'] == 'image' %}{{':'}}{% else %}{{': '}}{% endif %}{% for line in message['content'] %}{% if line['type'] == 'text' %}{{line['text']}}{% elif line['type'] == 'image' %}{{ '<image>' }}{% endif %}{% endfor %}<end_of_utterance>\n{% endfor %}{% if add_generation_prompt %}{{ 'Assistant:' }}{% endif %}"
|
98 |
+
tokenizer = AutoTokenizer.from_pretrained(base_model_id, use_fast=True)
|
99 |
+
tokenizer.chat_template = IDEFICS2_CHAT_TEMPLATE
|
100 |
+
processor.tokenizer = tokenizer
|
101 |
+
```
|
102 |
+
|
103 |
+
Simple inference example:
|
104 |
+
|
105 |
+
```
|
106 |
+
from transformers.image_utils import load_image
|
107 |
+
|
108 |
+
image = load_image("https://d2r55xnwy6nx47.cloudfront.net/uploads/2018/02/Ants_Lede1300.jpg")
|
109 |
+
|
110 |
+
# Create inputs
|
111 |
+
messages = [
|
112 |
+
{
|
113 |
+
"role": "user",
|
114 |
+
"content": [
|
115 |
+
{"type": "image"},
|
116 |
+
{"type": "text", "text": "What is shown in this image, and what is the relevance for materials design? Include a discussion of multi-agent AI."},
|
117 |
+
]
|
118 |
+
},
|
119 |
+
]
|
120 |
+
prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
|
121 |
+
|
122 |
+
# Get inputs using the processor
|
123 |
+
inputs = processor(text=prompt, images=[image], return_tensors="pt")
|
124 |
+
inputs = {k: v.to(DEVICE) for k, v in inputs.items()}
|
125 |
+
|
126 |
+
# Generate
|
127 |
+
generated_ids = model.generate(**inputs, max_new_tokens=500)
|
128 |
+
generated_texts = processor.batch_decode(generated_ids, skip_special_tokens=True)
|
129 |
+
|
130 |
+
print(generated_texts)
|
131 |
+
```
|
132 |
+
|
133 |
+
Next we provide a convenience function for inference. This function takes the model, processor, question, and images, along with messages and images objects for repeated chat-like interactions with the model.
|
134 |
+
|
135 |
+
```python
|
136 |
+
def ask_about_image (model, processor, question,
|
137 |
+
images_input=[],
|
138 |
+
verbatim=False,
|
139 |
+
temperature=0.1,
|
140 |
+
show_image=False,
|
141 |
+
system="You are a biomaterials scientist who responds accurately. ",
|
142 |
+
init_instr = "",
|
143 |
+
show_conversation=True,
|
144 |
+
max_new_tokens=256,
|
145 |
+
messages=[],
|
146 |
+
images=[],
|
147 |
+
use_Markdown=False,
|
148 |
+
):
|
149 |
+
|
150 |
+
|
151 |
+
query = question
|
152 |
+
images_input=ensure_list(images_input)
|
153 |
+
if len (images)==0:
|
154 |
+
if len (images_input)>0:
|
155 |
+
for image in tqdm (images_input) :
|
156 |
+
if is_url(image):
|
157 |
+
image= load_image(image)
|
158 |
+
images.append (image)
|
159 |
+
|
160 |
+
if show_image:
|
161 |
+
display ( image )
|
162 |
+
if len (messages)==0:
|
163 |
+
|
164 |
+
base_message = {
|
165 |
+
"role": "user",
|
166 |
+
"content": [
|
167 |
+
{"type": "text", "text": system + init_instr},
|
168 |
+
# Image messages will be added dynamically here
|
169 |
+
{"type": "text", "text": query}
|
170 |
+
]
|
171 |
+
}
|
172 |
+
|
173 |
+
# Ensure the images_input is a list
|
174 |
+
images_input = ensure_list(images_input)
|
175 |
+
|
176 |
+
# Add image messages dynamically
|
177 |
+
image_messages = [{"type": "image"} for _ in images_input]
|
178 |
+
base_message["content"][1:1] = image_messages # Insert image messages before the last text message
|
179 |
+
|
180 |
+
# Append the constructed message to messages list
|
181 |
+
messages.append(base_message)
|
182 |
+
|
183 |
+
else:
|
184 |
+
messages.append (
|
185 |
+
{
|
186 |
+
"role": "user",
|
187 |
+
"content": [
|
188 |
+
{"type": "text", "text": query
|
189 |
+
}
|
190 |
+
]
|
191 |
+
}
|
192 |
+
)
|
193 |
+
if verbatim:
|
194 |
+
print (messages)
|
195 |
+
|
196 |
+
text = processor.apply_chat_template(messages, add_generation_prompt=True)
|
197 |
+
inputs = processor(text=[text.strip()], images=images, return_tensors="pt", padding=True).to(DEVICE)
|
198 |
+
|
199 |
+
generated_ids = model.generate(**inputs, max_new_tokens=max_new_tokens, temperature=temperature, do_sample=True)
|
200 |
+
generated_texts = processor.batch_decode(generated_ids[:, inputs["input_ids"].size(1):], skip_special_tokens=True)
|
201 |
+
|
202 |
+
messages.append (
|
203 |
+
{
|
204 |
+
"role": "assistant",
|
205 |
+
"content": [ {"type": "text", "text": generated_texts[0]}, ]
|
206 |
+
}
|
207 |
+
)
|
208 |
+
formatted_conversation = format_conversation(messages, images)
|
209 |
+
|
210 |
+
# Display the formatted conversation, e.g. in Jupyter Notebook
|
211 |
+
if show_conversation:
|
212 |
+
|
213 |
+
if use_Markdown:
|
214 |
+
display(Markdown(formatted_conversation))
|
215 |
+
else:
|
216 |
+
display(HTML(formatted_conversation))
|
217 |
+
|
218 |
+
return generated_texts, messages, images
|
219 |
+
|
220 |
+
question = "What is shown in this image, and what is the relevance for materials design? Include a discussion of multi-agent AI."
|
221 |
+
|
222 |
+
url1 = "https://d2r55xnwy6nx47.cloudfront.net/uploads/2018/02/Ants_Lede1300.jpg"
|
223 |
+
|
224 |
+
response, messages,images= ask_about_image ( model, processor, question,
|
225 |
+
images_input=[url1,],
|
226 |
+
temperature=0.1,
|
227 |
+
system= '', init_instr='You carefully study the image, and respond accurately, but succinctly. Think step-by-step.\n\n',
|
228 |
+
show_conversation=True,
|
229 |
+
max_new_tokens=512, messages=[], images=[])
|
230 |
+
```
|
231 |
+
|
232 |
+
Sample output:
|
233 |
+
|
234 |
+
![image/png](https://cdn-uploads.huggingface.co/production/uploads/623ce1c6b66fedf374859fe7/5n6oRNHrfwHkBX0QertZp.png)
|
235 |
+
<small>Image by [Vaishakh Manohar](https://www.quantamagazine.org/the-simple-algorithm-that-ants-use-to-build-bridges-20180226/)</small>
|
236 |
+
|
237 |
+
<pre style="white-space: pre-wrap;">
|
238 |
+
The image depicts a group of ants moving in a coordinated manner to climb a vertical surface. This behavior is known as cooperative climbing and involves the use of multiple agents working together to achieve a common goal. The relevance for materials design lies in the potential application of multi-agent AI in developing new materials with improved properties through the collaboration of multiple agents.
|
239 |
+
</pre>
|
240 |
+
|
241 |
+
## Dataset generation
|
242 |
+
|
243 |
+
The schematic below shows a visualization of the approach to generate datasets for training the vision model. The extraction process employs advanced algorithms to accurately detect and separate images and their corresponding textual descriptions from complex PDF documents. It involves extracting images and captions from PDFs to create well-reasoned image-text pairs, utilizing large language models (LLMs) for natural language processing. These image-text pairs are then refined and validated through LLM-based NLP processing, ensuring high-quality and contextually relevant data for training.
|
244 |
+
|
245 |
+
The image below shows reproductions of two representative pages of the scientific article (here, Spivak, Buehler, et al., 2011), and how they are used to extract visual scientific data for training the Cephalo model.
|
246 |
+
|
247 |
+
![image/png](https://cdn-uploads.huggingface.co/production/uploads/623ce1c6b66fedf374859fe7/qHURSBRWEDgHy4o56escN.png)
|
248 |
+
|
249 |
+
# Further model optimizations
|
250 |
+
|
251 |
+
If your GPU allows, load and run inference in half precision (`torch.float16` or `torch.bfloat16`).
|
252 |
+
|
253 |
+
```diff
|
254 |
+
model = AutoModelForVision2Seq.from_pretrained(
|
255 |
+
"lamm-mit/Cephalo-Idefics-2-vision-8b-alpha",
|
256 |
+
+ torch_dtype=torch.float16,
|
257 |
+
).to(DEVICE)
|
258 |
+
```
|
259 |
+
|
260 |
+
**Vision encoder efficiency**
|
261 |
+
|
262 |
+
Given the high resolution supported, the vision part of the model can be memory hungry depending on your configuration. If you are GPU-memory-constrained, you can:
|
263 |
+
- **deactivate the image splitting.** To do so, add `do_image_splitting=False` when initializing the processor (`AutoProcessor.from_pretrained`). There are no changes required on the model side. Note that only the sft model has been trained with image splitting.
|
264 |
+
- **decrease the maximum image resolution.** To do so, add `size= {"longest_edge": 448, "shortest_edge": 378}` when initializing the processor (`AutoProcessor.from_pretrained`). In particular, the `longest_edge` value can be adapted to fit the need (the default value is `980`). We recommend using values that are multiples of 14. There are no changes required on the model side.
|
265 |
+
|
266 |
+
`do_image_splitting=True` is especially needed to boost performance on complex tasks where a very large image is used as input. The model was fine-tuned with image splitting turned on. For simple tasks, this argument can be safely set to `False`.
|
267 |
+
|
268 |
+
**Using Flash-attention 2 to speed up generation**
|
269 |
+
|
270 |
+
<details><summary>Click to expand.</summary>
|
271 |
+
|
272 |
+
Mke sure to install `flash-attn`. Refer to the [original repository of Flash Attention](https://github.com/Dao-AILab/flash-attention) for the package installation. Simply change the snippet above with:
|
273 |
+
|
274 |
+
```diff
|
275 |
+
model = AutoModelForVision2Seq.from_pretrained(
|
276 |
+
"lamm-mit/Cephalo-Idefics-2-vision-8b-alpha",
|
277 |
+
+ torch_dtype=torch.bfloat16,
|
278 |
+
+ _attn_implementation="flash_attention_2",
|
279 |
+
).to(DEVICE)
|
280 |
+
```
|
281 |
+
|
282 |
+
</details>
|
283 |
+
|
284 |
+
**4 bit quantization with bitsandbytes**
|
285 |
+
|
286 |
+
<details><summary>Click to expand.</summary>
|
287 |
+
It is possible to load Idefics2 in 4bits with `bitsandbytes`. Make sure that you have `accelerate` and `bitsandbytes` installed.
|
288 |
+
|
289 |
+
```diff
|
290 |
+
+ from transformers import BitsAndBytesConfig
|
291 |
+
|
292 |
+
quantization_config = BitsAndBytesConfig(
|
293 |
+
load_in_4bit=True,
|
294 |
+
bnb_4bit_quant_type="nf4",
|
295 |
+
bnb_4bit_use_double_quant=True,
|
296 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
297 |
+
)
|
298 |
+
model = AutoModelForVision2Seq.from_pretrained(
|
299 |
+
"lamm-mit/Cephalo-Idefics-2-vision-8b-alpha",
|
300 |
+
+ torch_dtype=torch.bfloat16,
|
301 |
+
+ quantization_config=quantization_config,
|
302 |
+
).to(DEVICE)
|
303 |
+
```
|
304 |
|
305 |
+
</details>
|
|
|
|
|
306 |
|
|
|
307 |
|
308 |
+
## Citation
|
309 |
|
310 |
+
Please cite as:
|
311 |
|
312 |
+
```bibtex
|
313 |
+
@article{Buehler_Cephalo_2024,
|
314 |
+
title = {Cephalo, a series of multi-modal vision-language models for bio-inspired materials and mechanics},
|
315 |
+
author = {M.J. Buehler},
|
316 |
+
journal = {},
|
317 |
+
year = {2024},
|
318 |
+
volume = {},
|
319 |
+
pages = {},
|
320 |
+
url = {}
|
321 |
+
}
|
322 |
+
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|