Clarify and add missing typical_p argument docstring. (#21095)

* Clarify and add missing typical_p docstring.

* Make the docstring easier to understand.

* Clarify typical_p docstring

Accept the suggestion by @stevhliu for paraphrasing the docstring.
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

* Use the same docstring as in GenerationConfig

Follow the suggestion suggested by @stevhliu in the pull request conversation.

* Fix docstring spacing.
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>

src/transformers/configuration_utils.py

@@ -144,6 +144,12 @@ class PretrainedConfig(PushToHubMixin):
         top_p (`float`, *optional*, defaults to 1):
             Value that will be used by default in the `generate` method of the model for `top_p`. If set to float < 1,
             only the most probable tokens with probabilities that add up to `top_p` or higher are kept for generation.
+        typical_p (`float`, *optional*, defaults to 1):
+            Local typicality measures how similar the conditional probability of predicting a target token next is to
+            the expected conditional probability of predicting a random token next, given the partial text already
+            generated. If set to float < 1, the smallest set of the most locally typical tokens with probabilities that
+            add up to `typical_p` or higher are kept for generation. See [this
+            paper](https://arxiv.org/pdf/2202.00666.pdf) for more details.
         repetition_penalty (`float`, *optional*, defaults to 1):
             Parameter for repetition penalty that will be used by default in the `generate` method of the model. 1.0
             means no penalty.

src/transformers/generation/configuration_utils.py

@@ -111,8 +111,11 @@ class GenerationConfig(PushToHubMixin):
             If set to float < 1, only the smallest set of most probable tokens with probabilities that add up to
             `top_p` or higher are kept for generation.
         typical_p (`float`, *optional*, defaults to 1.0):
-            The amount of probability mass from the original distribution to be considered in typical decoding. If set
-            to 1.0 it takes no effect. See [this paper](https://arxiv.org/pdf/2202.00666.pdf) for more details.
+            Local typicality measures how similar the conditional probability of predicting a target token next is to
+            the expected conditional probability of predicting a random token next, given the partial text already
+            generated. If set to float < 1, the smallest set of the most locally typical tokens with probabilities that
+            add up to `typical_p` or higher are kept for generation. See [this
+            paper](https://arxiv.org/pdf/2202.00666.pdf) for more details.
         diversity_penalty (`float`, *optional*, defaults to 0.0):
             This value is subtracted from a beam's score if it generates a token same as any beam from other group at a
             particular time. Note that `diversity_penalty` is only effective if `group beam search` is enabled.