lecture 11
NLP103:
Use ChatGPT in your app
3D content generation: NeRF
SeAts APp SEAtS ApP SEaTS APP
🧘
a
video with an interesting perspective to wake us up
after today's lecture:
-- Have ChatGPT on your phone
-- Basic introduction to the emerging 3D content generation technique NeRF
previously seen NLP (tokennization, sentiment analysis, etc.):
👶
the evolution pathway to ChatGPT
-- architecture improvement:
the "T" in GPT: transformer (attention mechanism)
-- upscaled data:
570GB of data obtained from various sources
-- upscaled model:
175 Billion parameters
-- upscaled computation resource:
~10,000 GPUs
etc.
today let's bring this gem into our phone (any ios app actually)
Example App code
here
referenced from
this article
API key ???
-- have a read on the OpenAI API document
here
-- have a read on how to get OpenAI API key
here
what's the model name for speech recognition?
1. generate a new OpenAI API key
here
2. paste that into the code
3. build
me demo 💅
To incorporate this to any other app:
1. Check the last section from this
article
2. See how the example code generates request(input) and handles response(output), both in JSON
Lil exercise, how to turn this app into an auto-summary app?
that is to,
input a paragraph to be summarised
output a summary
the response is slow 🥲:
-- there are
other language models that are faster
-- timeout parameter in line 94 from OpenAIConnector
A google colab playground:
DreamFields
parameters to play with (under "training settings"):
-- Epoch_num
-- clip_model (ViT/B-32 for less memory)
etc.
today we talked about:
- OpenAI API
- Incorporate ChatGPT to your app
- NeRF
Next:
Beyond generative AIs
Recap
Presentation brief
Mock multiple choices exam