Marc Green

icon location founding ai engineer @ tryusercue.com, 15 years soft eng, 11 years cybersec

Results 5 comments of Marc Green

add system prompt design to GPT model

> Out of curiosity, what would you like to put in the system prompt? Want to use the same pipeline across projects, with varying background context. The pipelines capture the...

Change tokenizer to use gpt-3-encoder for splitting

#3 adds gpt-3-encoder to sql2llm. references to the other tokenizer used, `tokenize_native` (via `natural.TreebankWordTokenizer()`), still exist in workflows/prompt.ts and workflows/embeddings.ts

Change tokenizer to use gpt-3-encoder for splitting

I agree, I think it makes sense to use this everywhere. Those two files I listed are the only other places I see any tokenizers being used in the project....

great work, needs better example

I updated the task to the task from the paper: ``` task_example = """ This SVG path element draws a: (A) circle (B) heptagon (C) hexagon (D) kite (E) line...

great work, needs better example

Presumably poor performance is related to the prompts used in this implementation. However, even using prompts that seem to more closely resemble the paper's (see: https://github.com/waszumteufel/autologic/blob/main/src/autologic/config.py), I am not not...