2.0.11

Jokes.

2.0.10

Fix PLAINFORMAT instruction format.

v2.0.9

Minor settings and prompt updates.

2.0.8

Remove filtering from some instructors, ensure accurate counts, misc fixes.

2.0.7

Misc fixes/improvements

2.0.6

Update the chain-of-thought prompt to di-bias position 1 for optimal solution.

2.0.5

Writing instructor

Creative writing task creation, with additional nuance such as allowing for tragic or sad endings, violence, etc.

Character/scenario/model card instructor

Allow creating simple, random model cards with attributes.

Concurrency improvements

Use async tasks to start generating from all instructors concurrently.

2.0.3

Add reWOO style planning instructor.

2.0.2

Fix regex to filter out refusals.

2.0.1

Massive overhaul:

• switch from multi-threaded to asyncio to reduce system load
• move from a single prompt to a series of instructors, each with their own customizable prompt/config
• experimental support for non-english language
• YAML based configuration instead of mostly CLI arguments
• currently available instructors:
  • agent/router
  • coding
  • contextual question answering
  • counterfactual contextual question answering (to lower hallucations when performing contextual QA)
  • chain-of-thought style prompts, with multiple possible solutions, ranking, and optimal selection
  • experiences, e.g. guided meditations or exploring a museum, etc.
  • general, i.e. random prompts without any seed values, wholly generated by the LLM
  • Orca, i.e. math or reasoning question with an added "ELI5"
  • riddles
  • trivia
  • wordgames, e.g. generating a list of words containing a string, starting with a string, etc.