If another file anywhere in `unit_tests` sets `langchain.verbose =
True`, it messes up all of the tests that check for no callbacks because
the `None` for verbose gets overridden by the global verbosity flag. By
explicitly setting it in unit tests, we bypass that potential issue
first pass at stdout callback
for the most part, went pretty smoothly. aside from the code here, here
are some comments/observations.
1.
should somehow default to stdouthandler so i dont have to do
```
from langchain.callbacks import get_callback_manager
from langchain.callbacks.stdout import StdOutCallbackHandler
get_callback_manager().add_handler(StdOutCallbackHandler())
```
2. I kept around the verbosity flag. 1) this is pretty important for
getting the stdout to look good for agents (and other things). 2) I
actually added this for LLM class since it didn't have it.
3. The only part that isn't basically perfectly moved over is the end of
the agent run. Here's a screenshot of the new stdout tracing
Noticing it is missing logging of the final thought, eg before this is
what it looked like
The reason its missing is that this was previously logged as part of
agent end (lines 205 and 206)
this is probably only relevant for the std out logger? any thoughts for
how to get it back in?
This PR has two contributions:
1. Add test for when stop token is found in middle of text
2. Add code coverage tooling and instructions
- Add pytest-cov via poetry
- Add necessary config files
- Add new make instruction for `coverage`
- Update README with coverage guidance
- Update minor README formatting/spelling
Co-authored-by: Hunter Gerlach <hunter@huntergerlach.com>
Add support for huggingface hub
I could not find a good way to enforce stop tokens over the huggingface
hub api - that needs to hopefully be cleaned up in the future
