The more efficient tokenizer is really interesting for non-English languages I wrote a little about how OpenAI's tokenizer penalizes languages like Spanish in this piece about how tokenizers work: https://
simonwillison.net/2023/Jun/8/gpt
-tokenizers/
…
LLMS
-
Tokenizer Efficiency and Language Bias in AI Models
By
–
-
OpenAI Enhances Fine-Tuning API with New Dashboards and Integrations
By
–
We're introducing new dashboards, metrics, and integrations in the fine-tuning API to give developers more control, and adding new ways to build custom models with OpenAI.
-
April Fine-Tuned Newsletter: LLM Serving Course and Updates
By
–
This month's Fine-Tuned newsletter showcases our new @deeplearningai short course "Efficiently Serving LLMs," recent seminars, webinars, and blog posts, and more!
-
Draft Model Size Considerations in AI Development
By
–
good point but the draft model is relatively small
-
Command R Plus Now Available in LLM CLI Tool
By
–
Command R+ (and Command R) are now available in my LLM CLI tool using the new llm-command-r plugin pipx install llm
llm install llm-command-r
llm keys set cohere llm -m command-r-plus "Say hello from Command R Plus" -
Cohere v5 SDK and LangChain Integration Package Launch
By
–
@cohere Integration Package Hot off the heels of Cohere's v5 sdk supporting tool calling and their new Command R+ model, we're excited to announce the `langchain-cohere` package, which exposes all of their integrations ranging from their chat models to model-specific
-
3Blue1Brown Visual Guide to Understanding Transformer Architecture
By
–
Visual introduction to Transformers! 3Blue1Brown released a video explaining the underlying basic cocepts of transformers with visual illustrations. "But what is a GPT? Visual intro to Transformers" https://
youtu.be/wjZofJX0v4M?si
=nFrtsqFGk_ADEAgi
… -
New Model Weights Available But Needs Claude 3 Comparison
By
–
Way too early to tell – I like that it's weights-available even though you can't use those commercially, it would need to prove itself against Claude 3
-
Speculative Decoding: 3-4x Faster Model Inference Performance
By
–
doesn't speculative decoding basically solve this problem already? (I think we can already run models 3-4x faster in practice by using a draft model and doing speculative decoding)
-
Cohere Command R Plus Model Card Details Released
By
–
Minimal details of the training data in the model card aside from a list of 23 languages (10 as a focus, 13 additional) that were covered https://
huggingface.co/CohereForAI/c4
ai-command-r-plus
…