Skip to content

armgit / dspy

Public Imported from GitHub
0

DSPy: The framework for programming—not prompting—language models

Python

Explained teacher settings better

Michael Ryan1ee5479b201b
diff --git a/dspy/teleprompt/mipro_optimizer_v2.py b/dspy/teleprompt/mipro_optimizer_v2.py
index 0c8457ff7..6188aeb52 100644
--- a/dspy/teleprompt/mipro_optimizer_v2.py
+++ b/dspy/teleprompt/mipro_optimizer_v2.py
@@ -48,8 +48,10 @@ eval_score = evaluate(compiled_prompt_opt, devset=evalset[:EVAL_NUM], **kwargs)
Note that this teleprompter takes in the following parameters:
* prompt_model: The model used for prompt generation. When unspecified, defaults to the model set in settings (i.e., dspy.settings.configure(lm=task_model)).
-* task_model: The model used for prompt generation. When unspecified, defaults to the model set in settings (i.e., dspy.settings.configure(lm=task_model)).
+* task_model: The model used for running your task. When unspecified, defaults to the model set in settings (i.e., dspy.settings.configure(lm=task_model)).
* teacher_settings: The settings used for the teacher model. When unspecified, defaults to the settings set in settings (i.e., dspy.settings.configure(lm=task_model)).
+ The teacher settings are used to generate the fewshot examples. This is the LLM/settings to use as a task model for the bootstrapping runs.
+ Typically you would want to use a model of equal or greater quality to your task model.
* metric: The task metric used for optimization.
* num_candidates: The number of new prompts and sets of fewshot examples to generate and evaluate. Default=10.
* init_temperature: The temperature used to generate new prompts. Higher roughly equals more creative. Default=1.0.