* Relax verify_first_message_correctness to accept any function call
* Also allow missing internal monologue if request_heartbeat
* Cleanup
* get instead of raw dict access
* Support attach in memgpt autogen agent
* Add docs example
* Add documentation, cleanup
* Relax verify_first_message_correctness to accept any function call
* Also allow missing internal monologue if request_heartbeat
* Cleanup
* get instead of raw dict access
- Removed repeating every user message like bein in debug mode
- Re-added the "dump" flag for the user message, to make it look nicer.
I may "reformat" other message too when dumping, but that was what
sticked out to me as unpleasant.
* add llamacpp server support
* use gbnf loader
* cleanup and warning about grammar when not using llama.cpp
* added memgpt-specific grammar file
* add grammar support to webui api calls
* black
* typo
* add koboldcpp support
* no more defaulting to webui, should error out instead
* fix grammar
* patch kobold (testing, now working) + cleanup log messages
Co-Authored-By: Drake-AI <drake-ai@users.noreply.github.com>
* I added a "/retry" command to retry for getting another answer.
- Implemented to pop messages until hitting the last user message. Then
extracting the users last message and sending it again. This will also
work with state files and after manually popping messages.
- Updated the README to include /retry
- Update the README for "pop" with parameter and changed default to 3 as
this will pop "function/assistant/user" which is the usual turn
around.
* disclaimer
---------
Co-authored-by: Charles Packer <packercharles@gmail.com>
* I added commands to shape the conversation:
`/rethink <text>` will change the internal dialog of the last assistant message.
`/rewrite <text>` will change the last answer of the assistant.
Both commands can be used to change how the conversation continues in
some pretty drastic and powerfull ways.
* remove magic numbers
* add disclaimer
---------
Co-authored-by: cpacker <packercharles@gmail.com>
* I made dump showing more message and added a count (the last x)
There seem to be some changes about the implementation so that the
current dump message helper functions do not show a lot of useful info.
I changed it so that you can `dump 5` (last 5 messages) and that it will
print user readable output. This lets you get some more understanding about
what is going on.
As some messages are still not shown I also show the index (reverse) of the
printed message, so one can see what to "pop" to reach a special point
without geting into the drumpraw.
* black
* patch
---------
Co-authored-by: Charles Packer <packercharles@gmail.com>
* strip '/' and use osp.join
* grepped for MEMGPT_DIR, found more places to replace '/'
* typo
* grep pass over filesep
---------
Co-authored-by: Vivian Fang <hi@vivi.sh>
* trying to patch summarize when running with local llms
* moved token magic numbers to constants, made special localllm exception class (TODO catch these for retry), fix summarize bug where it exits early if empty list
* missing file
* raise an exception on no-op summary
* changed summarization logic to walk forwards in list until fraction of tokens in buffer is reached
* added same diff to sync agent
* reverted default max tokens to 8k, cleanup + more error wrapping for better error messages that get caught on retry
* patch for web UI context limit error propogation, using best guess for what the web UI error message is
* add webui token length exception
* remove print
* make no wrapper warning only pop up once
* cleanup
* Add errors to other wrappers
---------
Co-authored-by: Vivian Fang <hi@vivi.sh>
