Add Model Context Protocol#4
Conversation
This reverts commit 5c52e0c.
|
@KnathanM something is wrong with the 'check for new stuff' part of the CI, could you take a look? |
|
@KnathanM thanks for fixing that! Looks like the test ran sucessfully. Please review at your leisure - @shihchengli and I have combed through it pretty well, so hopefully it doesn't need much. |
|
@KnathanM I have added the citation for the white paper describing the tool. This is good to go from our end. |
KnathanM
left a comment
There was a problem hiding this comment.
Yeah looks good. I'll leave a few questions, but we can merge this whenever.
Wait actually the Mac tests are failing. The moe test passes which is what I think gives the action a pass, so two questions are:
- What are the tests failing on mac
- Why does the CI not care that some of the tests with pytest are failing.
|
Here's a link to the failure: |
|
The GHA passes because each step's status is determined by the last command that was called. In the for loop of pytest calls, the last command is not pytest but actually the exit of the loop ( |
|
I haven't yet figured out why the actual test fails on Mac. |
|
@KnathanM I fixed the failing tests and addressed your comments as well - please re-review when possible! |
KnathanM
left a comment
There was a problem hiding this comment.
Thanks for finding those fixes.
Description
This PR add a Chemprop Model Context Protocol, which allows LLMs to call Chemprop training, prediction, fingerprinting, conversion, and hyperparameter optimization functions.
It is confirmed working with Chemprop 2.2.0 and 2.2.1
Questions
So far I have only added tests for
chemprop_trainsince the others are largely derivative of this one - we can add more, if we would like, but the fragile part of this code is the server and the interconnection which the current test proves is working.Checklist
Forgive the messy commit history, I accidentally pushed this to
mainand had to revert it.cc @shihchengli the co-author on this project