Sandeep Srinivasa
Sandeep Srinivasa
we use JSONObject in multiple places. If we can reduce amount of code using GSON/Jackson/Moshi - we should do that. Performance is not the top criteria here.
https://twitter.com/IR_oldie/status/1704464170446721199 check search relevance - llm/rag evaluation
https://github.com/stanfordnlp/CoreNLP caveat - https://news.ycombinator.com/item?id=14935005
 edgechains must have built in support to capture feedback for prompts/final answers (just like for token counts which we already do)
https://twitter.com/AlphaSignalAI/status/1703825582889263473 we should have a prompt router that detects questions like " summarize that document" and redirects it to chain of density summarization https://arxiv.org/pdf/2309.04269.pdf
https://arxiv.org/abs/2309.03409
https://twitter.com/AlexReibman/status/1704212821607645332 https://github.com/anthropics/anthropic-tokenizer-typescript https://github.com/anthropics/anthropic-tokenizer-typescript/blob/bd241051066ea37120f2898419e3fc8662fab280/index.ts#L3C1-L3C1
https://github.com/openai/openai-cookbook/blob/fd4e31bb000d4e86d5251d979ccff285160bc32d/examples/data/toy_chat_fine_tuning.jsonl this will help in finetuning.
building on top of @robincakeellis work in https://github.com/arakoodev/EdgeChains/issues/194 , we want to modularize the classes so i can use the classes from spring boot or android. Thankfully we already use...
https://github.com/orgs/community/discussions/14407 https://github.com/actions-cool/issues-helper/blob/5457ae8d7c1dc20597a753501d30183ed8043c8b/web/docs/index.md?plain=1#L38