Woopsie. This isn't exactly a workflow that the coprocess needs to support, but it's convenient for users.
The assertion is not valid in this case because we are doing 2 evals in a row. It's okay to not actually be eval-ing anything in the 2nd call. We could skip it. A better approach would be to have ChatTrajectory know when we haven't backtracked and therefore don't need to re-eval anything.
chat.cc:248: bool rendezllama::commit_to_context(llama_context*, ChatDisplay&, ChatTrajectory&, const ChatOptions&): Assertion chat_traj.context_tokencount < chat_traj.token_count() failed.
Woopsie. This isn't exactly a workflow that the coprocess needs to support, but it's convenient for users.
The assertion is not valid in this case because we are doing 2 evals in a row. It's okay to not actually be eval-ing anything in the 2nd call. We could skip it. A better approach would be to have
ChatTrajectoryknow when we haven't backtracked and therefore don't need to re-eval anything.