Linzhan/UniMate
Updated • 740 • 87
kon g
dolemole
·
AI & ML interests
None yet
Recent Activity
liked a model 7 days ago
Linzhan/UniMate liked a model about 1 year ago
IndexTeam/IndexTTS-2 new activity over 1 year ago
calcuis/chatterbox-gguf:Just wanted to say thanks CalcuisOrganizations
None yet
liked a
model 7 days ago
liked a
model about 1 year ago
Just wanted to say thanks Calcuis
😎👍 2
39
#1 opened over 1 year ago
by 
Winnougan
liked 2
models over 2 years ago
reacted to akhaliq's post with ❤️ over 2 years ago
Post
Simple and Scalable Strategies to Continually Pre-train Large Language Models
Simple and Scalable Strategies to Continually Pre-train Large Language Models (2403.08763)
Large language models (LLMs) are routinely pre-trained on billions of tokens, only to start the process over again once new data becomes available. A much more efficient solution is to continually pre-train these models, saving significant compute compared to re-training. However, the distribution shift induced by new data typically results in degraded performance on previous data or poor adaptation to the new data. In this work, we show that a simple and scalable combination of learning rate (LR) re-warming, LR re-decaying, and replay of previous data is sufficient to match the performance of fully re-training from scratch on all available data, as measured by final loss and language model (LM) evaluation benchmarks. Specifically, we show this for a weak but realistic distribution shift between two commonly used LLM pre-training datasets (EnglishrightarrowEnglish) and a stronger distribution shift (EnglishrightarrowGerman) at the 405M parameter model scale with large dataset sizes (hundreds of billions of tokens). Selecting the weak but realistic shift for larger-scale experiments, we also find that our continual learning strategies match the re-training baseline for a 10B parameter LLM. Our results demonstrate that LLMs can be successfully updated via simple and scalable continual learning strategies, matching the re-training baseline using only a fraction of the compute. Finally, inspired by previous work, we propose alternatives to the cosine learning rate schedule that help circumvent forgetting induced by LR re-warming and that are not bound to a fixed token budget.
Simple and Scalable Strategies to Continually Pre-train Large Language Models (2403.08763)
Large language models (LLMs) are routinely pre-trained on billions of tokens, only to start the process over again once new data becomes available. A much more efficient solution is to continually pre-train these models, saving significant compute compared to re-training. However, the distribution shift induced by new data typically results in degraded performance on previous data or poor adaptation to the new data. In this work, we show that a simple and scalable combination of learning rate (LR) re-warming, LR re-decaying, and replay of previous data is sufficient to match the performance of fully re-training from scratch on all available data, as measured by final loss and language model (LM) evaluation benchmarks. Specifically, we show this for a weak but realistic distribution shift between two commonly used LLM pre-training datasets (EnglishrightarrowEnglish) and a stronger distribution shift (EnglishrightarrowGerman) at the 405M parameter model scale with large dataset sizes (hundreds of billions of tokens). Selecting the weak but realistic shift for larger-scale experiments, we also find that our continual learning strategies match the re-training baseline for a 10B parameter LLM. Our results demonstrate that LLMs can be successfully updated via simple and scalable continual learning strategies, matching the re-training baseline using only a fraction of the compute. Finally, inspired by previous work, we propose alternatives to the cosine learning rate schedule that help circumvent forgetting induced by LR re-warming and that are not bound to a fixed token budget.
Can I use this in ComfyUI or Auto1111 SD WebUI?
4
#8 opened almost 3 years ago
by 
RageshAntony
upvoted 2
papers almost 3 years ago
HierSpeech++: Bridging the Gap between Semantic and Acoustic Representation of Speech by Hierarchical Variational Inference for Zero-shot Speech Synthesis
Paper • 2311.12454 • Published • 29
CoMoSpeech: One-Step Speech and Singing Voice Synthesis via Consistency Model
Paper • 2305.06908 • Published • 7
liked a
model almost 3 years ago
upvoted 2
papers almost 3 years ago
updated a
collection almost 3 years ago
upvoted a collection about 3 years ago
updated a
collection about 3 years ago
liked 2
models about 3 years ago
liked a
dataset about 3 years ago