You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the lesson video at 5:25 you told that when we repeat the execution from the specific checkpoint then it will be very fast because no execution is performed because the graph knows that it was executed before.
But this sounds strange because if we run this code in the notebook and check the latency, answers and langsmith traces, we see that both replaying and forking will start all the steps after specific checkpoint we choose at this time.
This can be confirmed also with Time Travel section from Quick Start of the LangGraph documentation, where we also do not change the input for the specific checkpoint, and make steps similar to repeating, and we also see that the subsequent steps are executed (at least checking that the final answers for the original and repeated executions are different).
Could you please explain do I understand correctly your statement about "repeating" or there is some difference between behavior of the environment in the video and in the notebook with the latest packages?
I found multiple inconsistencies in the outputs for Lesson 3 in the same Module 3, and probably there were some changes in the libraries since video was recorded, however behavior of the Quick Start tutorial for LangGraph documentation is consistent between versions (including execution results in your documentation example) and always run the subsequent steps, not just replaying existing steps.
Thank you.
The text was updated successfully, but these errors were encountered:
labdmitriy
changed the title
Replay from checkout will run subsequent steps, not just replay it (Module 3 - Lesson 5: Time Travel)
Replay from checkpoint will run subsequent steps, not just replay it (Module 3 - Lesson 5: Time Travel)
Oct 9, 2024
Hi @rlancemartin,
In the lesson video at 5:25 you told that when we repeat the execution from the specific checkpoint then it will be very fast because no execution is performed because the graph knows that it was executed before.
But this sounds strange because if we run this code in the notebook and check the latency, answers and langsmith traces, we see that both replaying and forking will start all the steps after specific checkpoint we choose at this time.
This can be confirmed also with Time Travel section from Quick Start of the LangGraph documentation, where we also do not change the input for the specific checkpoint, and make steps similar to repeating, and we also see that the subsequent steps are executed (at least checking that the final answers for the original and repeated executions are different).
Could you please explain do I understand correctly your statement about "repeating" or there is some difference between behavior of the environment in the video and in the notebook with the latest packages?
I found multiple inconsistencies in the outputs for Lesson 3 in the same Module 3, and probably there were some changes in the libraries since video was recorded, however behavior of the Quick Start tutorial for LangGraph documentation is consistent between versions (including execution results in your documentation example) and always run the subsequent steps, not just replaying existing steps.
Thank you.
The text was updated successfully, but these errors were encountered: