Of course, when using an open source model (Vicuna, Dolly) that model name is not in the tiktoken wrapper Previously, this would raise an error We've now updated this to: (1) fallback to a default tokenizer, (2) allow users to specify model name to use for tiktoken
LLMS
-
LangChain PR improves open source model experimentation
By
–
PR for this here: https://
github.com/hwchase17/lang
chain/commit/e9877ea8b1d301efafb7d2fda5f9105b005774b8
… Shout out to Yuze Ma for highlighting this limitation at a hackathon this weekend! Hopefully this should make it easier to experiment with open source models -
LangChain API Base Configuration and Token Management
By
–
It's always been possible to change the API base in the LangChain model wrappers However, the model wrappers do a lot of handy things like token management, which rely on tokenizers that by default we look up in tiktoken We look this up based on model name
-
Open Source Model APIs Now OpenAI-Compatible via LangChain
By
–
Many startups focused on serving open source models have exposed APIs that are interoperable with OpenAI This makes it easy for anyone to try them out by just changing the API base We added better support for this in @langchain in the most recent release (h/t Yuze Ma)
-
PyTorch dominance in latest LLM and vision transformer research
By
–
That's totally true, but I think the problem is once you want to adopt or use more of the recent (research) stuff and/or need more customization. E.g., all the latest LLM and vision transformer is basically exclusively PyTorch.
-
Live Panel on Future of LLMs with MosaicML Chief Scientist
By
–
Join us this Wednesday for a live panel discussion with @mosaicml
’s Chief Scientist @jefrankle and our CEO @amasad to learn more about the future of LLMs. Our VP of AI @pirroh will be facilitating this illuminating conversation. -
MosaicML Launches Advanced LLM Model for Developers
By
–
Congratulations to the team @mosaicml for launching this exciting new model! Advanced LLMs like MosaicML’s and our own are changing the game for developers everywhere. How did we get here and what comes next?
-
Contrastive Search vs Greedy Search for LLM Generation
By
–
Yeh! Would deffo recommend contrastive search. It works quite well (however, over generates sometime ) Greedy search simply results in lost context.
-
Fine-tuned Models Performance on Multilingual Translation Tasks
By
–
Agreed! From my experience, it works well on languages that were abundant in the train set and also had lang -> en translate pairs. I’ll run some experiments to see how well do fine-tuned models perform. Quite lovely to see ya already using this
-
Testing Model Performance on Untrained Tasks: Benchmarking Results
By
–
Quick caveat: It's a task that the model isn't explicitly trained on, so test it thoroughly before us We'll run more benchmarks this week and share our findings soon! Help us test it better? 🙂