LLMS

Text-davinci-002 RLHF tuning clue in InstructGPT announcement footnote

By

–

30 November 2022 18h09

As far as I can tell, the only clue that that text‑davinci‑002 wasn't RLHF-tuned was this footnote in the InstructGPT announcement: https://
openai.com/blog/instructi
on-following/#fn2
…

Is text-davinci-002 RLHF-tuned? A footnote suggests otherwise.

By

–

30 November 2022 18h09

As far as I can tell, the only clue that that text‑davinci‑002 wasn't RLHF-tuned was this footnote in the InstructGPT announcement: https://
openai.com/blog/instructi
on-following/#fn2
…

text-davinci-003 is first RLHF model, not 002

By

–

30 November 2022 18h03

Also, it seems that the new text‑davinci‑003 is actually the *first* released model to incorporate RLHF tuning, using the PPO algorithm. I've been misdescribing text‑davinci‑002 as an RLHF model for the past year.

→ View original post on X — @cerebras

Sparse GPT-3 Models Achieve 3x FLOP Reduction on CS-2

By

@cerebras

–

30 November 2022 18h00

We trained unstructured sparse 1.3B GPT-3 models on CS-2 systems and demonstrated how we achieve competitive results at a fraction of the inference FLOPs – our 83.8% sparse model achieved a 3x reduction in FLOPs at matching performance Learn more here:

→ View original post on X — @langchain

LangChain 0.0.25: Bug Fixes and Feature Improvements

By

@langchain

–

30 November 2022 17h02

LangChain version 0.0.25 Bug fix for the semantic similarity example selector (@AkashSamant4)
Support SQL statements that return no results (@andrewgleave – 1st PR! Welcome!!)
Tidying up SerpAPI and PALChain
Adding examples of demos that use LangChain

→ View original post on X — @montreal_ai

BLOOM: 176B Parameter Open-Access Multilingual Language Model

By

@montreal_ai

–

30 November 2022 16h16

BLOOM: A 176B-Parameter Open-Access Multilingual Language Model Le Scao et al.: https://
arxiv.org/abs/2211.05100 #ArtificialIntelligence #DeepLearning #MachineLearning

Text-davinci-002 derived from code-davinci-002 with code training

By

–

30 November 2022 16h00

News to me that text-davinci-002 derives from code-davinci-002. I imagined the code model had code training the text one lacked. It looks like they both have code training, but instruction tuning interferes with code completion use cases maybe.

OpenAI retroactively updates training methods for GPT-3.5 models

By

–

30 November 2022 15h39

OpenAI update on training methods for “GPT‑3.5” models (retroactively including text‑davinci‑002): https://
beta.openai.com/docs/model-ind
ex-for-researchers
…