-
bigscience
Central place for the engineering/scaling WG: documentation, SLURM scripts and logs, compute environment and data.
-
InfluxDB
Power Real-Time Data Analytics at Scale. Get real-time insights from all types of time series data with InfluxDB. Ingest, query, and analyze billions of data points in real-time with unbounded cardinality.
-
t-zero
Reproduce results and replicate training fo T0 (Multitask Prompted Training Enables Zero-Shot Task Generalization)
So far, in the architecture & scaling group, we have been benchmarking models mainly with the EleutherAI Language Model Evaluation Harness. In cooperation with the multilingual group, we have also considered XNLI for multilingual evaluation. You can read more about our work in our paper.
Model architecture and a blog post on decisions on architecture, size, shape, and pretraining duration
And if you want to see int4 - please vote for it here: https://github.com/pytorch/pytorch/issues/74627 - unless the community votes I don't see it'll happen any time soon.
You have T0 that came from BigScience as well: https://github.com/bigscience-workshop/t-zero / https://huggingface.co/bigscience/T0
Related posts
-
[D] Keras 3.0 Announcement: Keras for TensorFlow, JAX, and PyTorch
-
Ask HN: What AI developer tools do you wish you'd discovered sooner?
-
[Guide] DreamBooth Training with ShivamShrirao's Repo on Windows Locally
-
[D] My experience with running PyTorch on the M1 GPU
-
Julia: faster than Fortran, cleaner than Numpy