Comments (2)
Yeah sorry, this would only be used for continued pretraining. I forgot to mention that.
from litgpt.
That's not a bad idea and will be more robust for different dataset sizes. I wonder (and I don't know the answer) though if there is a minimum number of warm up step that should always be done, and a max number of steps that shouldn't be exceeded.
For example if we use 0.05 and pretrain on 3T tokens, that's 150 billion warmup steps, which is a bit large :D
For reference, here are the number of steps for popular LLMs (taken from the OLMo paper)
![Screenshot 2024-04-02 at 1 43 50 PM](https://private-user-images.githubusercontent.com/5618407/318925389-e6be75a7-7a28-4b9e-88be-870e284b8e6b.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.5XSdfWw9y8hAIK_N-n-9fJ-RUL2pCYEvQevfihb48Qo)
from litgpt.
Related Issues (20)
- test_tinyllama issue with LitData and `iterate_over_all` HOT 2
- Remove old and unused LLMs
- Pretraining example from readme fails in Colab HOT 3
- Streamline LitGPT API HOT 7
- Redundancy? HOT 2
- support for qwen2 and baichuan
- 'Phi-3-mini-4k-instruct' is not a supported config name HOT 1
- Continue pre-training got RuntimeError: Failed processing /tmp/data HOT 4
- prompt_style HOT 4
- Lora recipes use lots of memory because of not wrapping parameters with gradient in separate FSDP unit HOT 2
- how to pretrain llama2? HOT 4
- Python API
- Stream option HOT 3
- Resolve output characters garbled HOT 4
- Continually pretrained Llama2-7B-hf model inference is not working on 16GB GPU machine HOT 5
- how to pretrain llama2 in custom data? HOT 1
- Is there any best practice for using litdata to load custom data for pretraining? HOT 1
- performing continuous pretraining and then finetuning causes error HOT 3
- pretrain custom dataset gpu memory oom
- Create new CI API key HOT 1
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 litgpt.