- added mussing types
- fixed typos in react imports

now

- swaped "Question" from prompt with "New Input", works better with llama-SFT-7E3
- seed -> 43
- similarity coef for tool selection reduced

currently testing: llama-sft-7-e3 and small fixes
- small prompt composition tweaks
- fix for chat memory deletion
- tweaking of sampling parameters
- increased limit of request result truncation

- tweaked sampling settings for plugin usage
- tweaked prompting templage
- fix for tool extraction function
- fix in final prompt when plugin is not used

…fig json preview

- updated main instruction prompts
- updated opeanapi parsers, with some basic support for post requests,
  and fixes for unsafe accessing of elements
- updated tools composition so now it is kind of usable for post
  requests
- removed unnecessary imports of Tuple, List etc...
- updated fetching of plugins to use aiohttp
- fix for frontend part, so now when user switches between presets,
  plugin config is preserved

… sampling settings to the UI with locking functionality

…orks much

better with current RLHF llama based models
- Implemented naive mechanism for fixing LLM generated JSON action input data
- Implemented truncation of the prompt template sequences from the final generation
to help prevention of the prompt leaks that mainly occurs when using RLHF llama
models or with any llama model, when truncate_prompt kicks and we end up loosing
beginning of the final composed prompt
- Removed included calculator plugin
- Various small fixes for UI based
- In RenderedMarkdown now images are supported, when appropriate input prop is provided

deleted package-lock.json in the root of the repo

- fixed conflicts with frontend part of the app