oellm_progress / README.md
geoalgo's picture
Update README.md
d19d622 verified
|
Raw History Blame Contribute Delete
920 Bytes

A newer version of the Gradio SDK is available: 6.29.1

Upgrade
metadata
title: OpenEuroLLM Training Progress
emoji: 📊
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.6.0
python_version: '3.12'
app_file: app.py
pinned: true
short_description: Live pretraining progress for OpenEuroLLM models

OpenEuroLLM Pretraining Progress

A live progress bar for the OpenEuroLLM 9B pretraining run: tokens seen vs. tokens to be processed.

Data is refreshed daily by the oellm-progress watcher in oellm-monitoring, which runs on an always-on local machine. It SSHes into Leonardo over a persistent mux, reads the latest training iteration from the Megatron stdout log, computes

tokens_seen  = iteration   * global_batch_size * seq_length
tokens_total = train_iters * global_batch_size * seq_length   (or a fixed token budget)

and uploads progress.json + progress_history.csv here. The Space only renders them.