Victor Dibia

icon indicating a website link http://victordibia.com

icon company Microsoft Corporation icon location Research Software Engineer. Interested in the intersection of Applied AI and HCI.

Results 218 comments of Victor Dibia

Is there a way to change the "Number of Images" setting to generate 1 image at a time instead of all at the same time?

This is an excellent suggestion. I'll explore an implementation

CUDA out of memory

Thanks for the feedback. There is a chance I should be attempting to flush memory after each generation. I will look into this. Conda shold not be a factor.

Regularization Loss Penalty

Hi @preyasgarg, Thanks for the note. Can you provide more information or context? How much improvement might this penalty provide and expected impact on generated images? Also, a PR would...

Can lida.summarize( ) ,input a long string instead of local file ?

Hi @nuaabuaa07 , Can you describe the use case you intend to support? E.g. what is the input and expected output? Currently, LIDA is optimized for tabular data, but I...

Dataset Finder (Data Discovery Tool)

Thanks Aiden. I am leaning more towards supporting **_discovery_** of data as opposed to hosting data (we probably can assume the user is able to do this already). I updated...

Error: name 'tiktoken' is not defined

are you able to run `pip install lida` without any errors? I typically suggest using a fresh python environment e.g. via conda.

'hf' model "Mixtral-8x7b" unable to load.

The mixtral models have not been tested with lida/llmx. I would recommend the following. - load the model using a tool like [vllm](https://docs.vllm.ai/en/latest/getting_started/quickstart.html#openai-compatible-server) which supports mixtral - vllm provides an...

Wrong expected type in class `Goal` from lida/datamodel.py

@trojrobert , any chance you want to open a PR for this?

The model did not return a valid JSON object

Hi, This is a known issue tends to arise with smaller models, also mentioned in #27 We are currently running some experiments to see which small model provides decent performance...

Methods to load the local downloaded LLM model

What quantized models are you interested in using? In general, as long as you can spin up an openai compliant webserver endpoint from your model, you can integrate it into...