TeachableAgent by rickyloynd-microsoft · Pull Request #278 · microsoft/autogen

rickyloynd-microsoft · 2023-10-18T00:38:31Z

Why are these changes needed?

TeachableAgent allows users to teach their assistant facts, preferences, and tasks unrelated to code generation.

Conversational assistants based on LLMs can remember the current chat with the user, and can even demonstrate in-context learning of things that the user teaches the assistant during the chat. But these memories and learnings are lost once the chat is over, or when a single chat grows too long. In subsequent chats, the user is forced to repeat any necessary instructions over and over.

TeachableAgent addresses these limitations by persisting user teachings across chat boundaries in long-term memory (a vector database). Memory is saved to disk at the end of each chat, then loaded from disk at the start of the next. Instead of copying all of memory into the context window, which would eat up valuable space, individual memories (called memos) are retrieved into context as needed. This allows the user to teach frequently used facts, preferences and skills to the agent just once, and have the agent remember them in later chats.

Related issue number

None

Checks

I've included any doc changes needed for https://microsoft.github.io/autogen/. See https://microsoft.github.io/autogen/docs/Contribute#documentation to build and test documentation locally.
I've added tests (if relevant) corresponding to the changes introduced in this PR.
I've made sure all auto checks have passed.

gagb · 2023-10-21T00:43:27Z

@rickyloynd-microsoft, minor comments that could be addressed in a different PR.

learn_from_user_feedback can be a long running job-- makes many API calls. Some notion of logging/progress bar would be helpful
learn_from_user_feedback could return/yield diff in learning from that call

What do you think?

rickyloynd-microsoft · 2023-10-21T00:45:17Z

@rickyloynd-microsoft, minor comments that could be addressed in a different PR.

learn_from_user_feedback can be a long running job-- makes many API calls. Some notion of logging/progress bar would be helpful

learn_from_user_feedback could return/yield diff in learning from that call

What do you think?

I like it. Probably in a new PR.

Replace agent with agent teachable_agent as recommended.

qingyun-wu · 2023-10-21T01:05:26Z

@rickyloynd-microsoft, minor comments that could be addressed in a different PR.

learn_from_user_feedback can be a long running job-- makes many API calls. Some notion of logging/progress bar would be helpful

learn_from_user_feedback could return/yield diff in learning from that call

What do you think?

I like it. Probably in a new PR.

Created a PR #329 @rickyloynd-microsoft @gagb

sonichi · 2023-10-21T01:05:38Z

@qingyun-wu We need to run OpenAI test before merging this PR. Please confirm that you understand when this is required or let me know if you have questions.

sonichi · 2023-10-21T01:06:01Z

@rickyloynd-microsoft, minor comments that could be addressed in a different PR.

learn_from_user_feedback can be a long running job-- makes many API calls. Some notion of logging/progress bar would be helpful

learn_from_user_feedback could return/yield diff in learning from that call

What do you think?

I like it. Probably in a new PR.

Created a PR #329 @rickyloynd-microsoft @gagb

You mean issue.

qingyun-wu · 2023-10-21T01:06:39Z

@rickyloynd-microsoft, minor comments that could be addressed in a different PR.

learn_from_user_feedback can be a long running job-- makes many API calls. Some notion of logging/progress bar would be helpful

learn_from_user_feedback could return/yield diff in learning from that call

What do you think?

I like it. Probably in a new PR.

Created a PR #329 @rickyloynd-microsoft @gagb

You mean issue.

Yes created an issue #329 @rickyloynd-microsoft @gagb

rickyloynd-microsoft · 2023-10-21T01:08:27Z

@qingyun-wu We need to run OpenAI test before merging this PR. Please confirm that you understand when this is required or let me know if you have questions.

I've got the latest changes ready that @sonichi and @qingyun-wu requested. Can I push those now? Those changes are to test on just one version of python, and to change agent to teachable_agent in places.

qingyun-wu · 2023-10-21T01:08:56Z

@qingyun-wu We need to run OpenAI test before merging this PR. Please confirm that you understand when this is required or let me know if you have questions.

You are right. We should run OpenAI test when there are new tests added. Is this understanding correct?

sonichi · 2023-10-21T01:13:13Z

@qingyun-wu We need to run OpenAI test before merging this PR. Please confirm that you understand when this is required or let me know if you have questions.

You are right. We should run OpenAI test when there are new tests added. Is this understanding correct?

Incorrect. Not all tests require openai, and sometimes openai test needs to be run even when no new test is added.
Also, don't run it unless it's the blocker of merging.

rickyloynd-microsoft added 30 commits September 20, 2023 17:04

Initial commit.

93ee755

Disable LLM response caching.

4d2ccf6

Merge branch 'main' into ricky

1c61c73

Add teachability option to setup.py

fa033c0

Modify test to use OAI_CONFIG_LIST as suggested in the docs.

165d5ce

Expand unit test.

295f9e0

Complete unit test.

515f2d9

Merge branch 'main' into ricky

07cc364

Add filter_dict

3443e57

details

dfe47a5

AnalysisAgent

c61d69c

details

748ffcd

Merge branch 'main' into ricky

1e6e889

More documentation and debug output.

25faf14

Support retrieval of any number of relevant memos, including zero.

66f4c61

More robust analysis separator.

e796832

cleanup

b31141f

teach_config

18387a7

refactoring

0f49f0d

For robustness, allow more flexibility on memo storage and retrieval.

09fde25

de-dupe the retrieved memos.

adac582

Simplify AnalysisAgent. The unit tests now pass with gpt-3.5

94d8ba0

comments

30503bf

Add a verbosity level to control analyzer messages.

4d2cedd

refactoring

8a15d0a

comments

4806eef

Persist memory on disk.

86e9e04

cleanup

60bd4c2

Merge branch 'main' into ricky

5f0075b

Use markdown to format retrieved memos.

b0ee207

skzhang1 approved these changes Oct 20, 2023

View reviewed changes

Merge branch 'main' into ricky

e13d512

sonichi had a problem deploying to openai October 21, 2023 00:12 — with GitHub Actions Error

sonichi had a problem deploying to openai October 21, 2023 00:12 — with GitHub Actions Failure

sonichi temporarily deployed to openai October 21, 2023 00:12 — with GitHub Actions Inactive

qingyun-wu added this pull request to the merge queue Oct 21, 2023

sonichi removed this pull request from the merge queue due to a manual request Oct 21, 2023

qingyun-wu mentioned this pull request Oct 21, 2023

Enhancements to TeachableAgent.learn_from_user_feedback #329

Closed

rickyloynd-microsoft added 2 commits October 20, 2023 18:04

Test on just python 3.10.

2745276

Replace agent with agent teachable_agent as recommended.

Merge branch 'ricky' of https://github.com/microsoft/autogen into ricky

de17a4e

Test on python 3.9 instead of 3.10.

ff4a369

rickyloynd-microsoft had a problem deploying to openai October 21, 2023 01:23 — with GitHub Actions Failure

sonichi reviewed Oct 21, 2023

View reviewed changes

Comment thread autogen/agentchat/contrib/teachable_agent.py Outdated

Remove space from name -> teachableagent

f962f45

rickyloynd-microsoft temporarily deployed to openai October 21, 2023 01:42 — with GitHub Actions Inactive

sonichi added this pull request to the merge queue Oct 21, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

TeachableAgent#278

TeachableAgent#278
sonichi merged 66 commits into
mainfrom
ricky

rickyloynd-microsoft commented Oct 18, 2023 •

edited

Loading

Uh oh!

gagb commented Oct 21, 2023 •

edited

Loading

Uh oh!

rickyloynd-microsoft commented Oct 21, 2023

Uh oh!

Uh oh!

qingyun-wu commented Oct 21, 2023

Uh oh!

sonichi commented Oct 21, 2023

Uh oh!

sonichi commented Oct 21, 2023

Uh oh!

qingyun-wu commented Oct 21, 2023

Uh oh!

rickyloynd-microsoft commented Oct 21, 2023 •

edited

Loading

Uh oh!

qingyun-wu commented Oct 21, 2023

Uh oh!

sonichi commented Oct 21, 2023

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants

Conversation

rickyloynd-microsoft commented Oct 18, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why are these changes needed?

Related issue number

Checks

Uh oh!

gagb commented Oct 21, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rickyloynd-microsoft commented Oct 21, 2023

Uh oh!

Uh oh!

qingyun-wu commented Oct 21, 2023

Uh oh!

sonichi commented Oct 21, 2023

Uh oh!

sonichi commented Oct 21, 2023

Uh oh!

qingyun-wu commented Oct 21, 2023

Uh oh!

rickyloynd-microsoft commented Oct 21, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

qingyun-wu commented Oct 21, 2023

Uh oh!

sonichi commented Oct 21, 2023

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants

rickyloynd-microsoft commented Oct 18, 2023 •

edited

Loading

gagb commented Oct 21, 2023 •

edited

Loading

rickyloynd-microsoft commented Oct 21, 2023 •

edited

Loading