Comments (5)
How about opening some layers or enabling the finetuning some parameters of the original LISA, do you think it would be doable? @X-Lai
from lisa.
There is only a single finetuning state. The reasoning segmentation dataset is directly involved with other datasets like the semantic / referring segmentation datasets.
from lisa.
There is only a single finetuning state. The reasoning segmentation dataset is directly involved with other datasets like the semantic / referring segmentation datasets.
emm,I want to employ the LISA to conduct the image-caption task, whose generated caption is more detailed than other model. But there is a domain difference between the train dataset and industrial dataset, so I wanna finetune LISA to apply to industrial domain.
from lisa.
That sounds great! Currently, we do not plan to release the code to further fine-tuning LISA. I think you could try to adopt the current hybrid-training paradigm and use your customized dataset together with other datasets.
You can add your customized data into the current training pipeline by simply configuring the dataset
argument. As you want to use VQA data to conduct the image-caption task, I think you can directly replace the current llava-instruct-150k
dataset with yours by simply configuring the vqa_data
argument in train_ds.py
.
from lisa.
That sounds great! Currently, we do not plan to release the code to further fine-tuning LISA. I think you could try to adopt the current hybrid-training paradigm and use your customized dataset together with other datasets.
You can add your customized data into the current training pipeline by simply configuring the
dataset
argument. As you want to use VQA data to conduct the image-caption task, I think you can directly replace the currentllava-instruct-150k
dataset with yours by simply configuring thevqa_data
argument intrain_ds.py
.
Your model is so excellent! I will try your suggestion, thank you!
from lisa.
Related Issues (20)
- Comparison with GroundedDINO HOT 1
- Inference speed on refcocog dataset
- Is LISA model on Hugginface? HOT 2
- Token indices sequence length is longer than the specified maximum sequence length for this model (565 > 512). Running this sequence through the model will result in indexing errors HOT 1
- How to generate ReasonSeg Dataset
- Why apply the causal mask to image tokens in the attention operations of LLM?
- How many samples were iterated in total? HOT 1
- LISA model_forward bug, only occurs when inference. HOT 2
- Code error about scheduler in train_ds.py
- Incorect online demo link
- Can't reproduce the results of lisa-7b HOT 4
- datasets download links
- flash_attn HOT 2
- LISA-7B-v1&& LISA-7B-v1-explanatory
- Finetuning Projection Layer or Not
- How to train my datesets?
- step vs epoch
- IndexError: The shape of the mask [6, 440] at index 1 does not match the shape of the indexed tensor [6, 760, 256] at index 1
- How to evaluate on RefCOCO datasets ?
- Supplementary material of the accepted paper
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from lisa.