The previous version was built on guidance 0.0.64, the older version built with handlebars-style syntax. However, this meant that we had to use an old version of OpenAI. This PR refactors BlendSQL to use the newer 0.1.0 version of guidance.
Reformatted _programs.py to use guidance>=0.1.0 syntax
Found in blendsql/_programs.py, each LLM ingredient now has a program represented as a Python script. This is cleaner imo, since we can use context handlers instead of rewriting completion/chat style prompts with the handlebar syntax
Restructured llms folder
I moved them to blendsql/llms, and created a remote/local subdirectories
num_prompt_tokens is now updated with each predict call
The previous version was built on guidance 0.0.64, the older version built with handlebars-style syntax. However, this meant that we had to use an old version of OpenAI. This PR refactors BlendSQL to use the newer 0.1.0 version of guidance.
_programs.py
to useguidance>=0.1.0
syntaxllms
folderremote
/local
subdirectoriesnum_prompt_tokens
is now updated with eachpredict
callAzureOpenaiLLM
class was added