Context Copy editing in Q4 2024 to update/refresh entire Prompt Engineering roadmap (1st/last edit: May 2023)
Issue: Under LLM Settings, sub topic Hyperparameters is blank, null, with no content/links.
Constraint: Current guidelines seek a concise single paragraph.
Backstory: In refreshing and researching LLM settings / hyperparameters, it was found / is obvious that Temperature and Top P are not the only ones. There are more, as follows:
Top K (similar, not as available) Top P
Stop Sequences
Frequency Penalty
Presence Penalty.
These are mostly available in OpenAI and or Google AI Studio, and less so on Anthropic, though by consensus, these are the other hyperparameters that could be exposed by direct API parameters (to be verified), when they may not be so readily available via Playground/Console UIs.
Condensing all of these into one paragraph (if inserted into one node only) will mean a loss of fidelity and meaning of their relative importance and usefulness to beginners. I.e. too compressed, and overly summarised.
Request:
Remove current node Hyperparameters under LLM settings
Add four nodes as above.
Content copy of each of these can be provided on request (single concise paragraphs, with links), after creation.
Links
Rational: Spliting a larger collective node / content into smaller discrete nodes / content improve readability, access and usefulness.
Roadmap URL
https://draw.roadmap.sh/6716b80e791f57dd60650121
Suggestions
Context Copy editing in Q4 2024 to update/refresh entire Prompt Engineering roadmap (1st/last edit: May 2023)
Issue: Under LLM Settings, sub topic
Hyperparameters
is blank, null, with no content/links. Constraint: Current guidelines seek a concise single paragraph.Backstory: In refreshing and researching LLM settings / hyperparameters, it was found / is obvious that Temperature and Top P are not the only ones. There are more, as follows:
These are mostly available in OpenAI and or Google AI Studio, and less so on Anthropic, though by consensus, these are the other hyperparameters that could be exposed by direct API parameters (to be verified), when they may not be so readily available via Playground/Console UIs.
Condensing all of these into one paragraph (if inserted into one node only) will mean a loss of fidelity and meaning of their relative importance and usefulness to beginners. I.e. too compressed, and overly summarised.
Request:
Hyperparameters
underLLM settings
Rational: Spliting a larger collective node / content into smaller discrete nodes / content improve readability, access and usefulness.