@@ -49,7 +49,7 @@ def create(
4949 * ,
5050 model : Literal ["kanon-2-enricher" ],
5151 texts : Union [SequenceNotStr [str ], str ],
52- overflow_strategy : Optional [Literal ["auto" , "drop_end" , "chunk" ]] | Omit = omit ,
52+ overflow_strategy : Optional [Literal ["auto" , "drop_end" ]] | Omit = omit ,
5353 # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
5454 # The extra values given here take precedence over values defined on the client or passed to this method.
5555 extra_headers : Headers | None = None ,
@@ -71,15 +71,9 @@ def create(
7171
7272 overflow_strategy: The strategy for handling content exceeding the model's maximum input length.
7373
74- `auto`, which is the recommended setting, currently behaves the same as `chunk`,
75- which intelligently breaks the input up into smaller chunks and then stitches
76- the results back together into a single prediction. In the future `auto` may
77- implement even more sophisticated strategies for handling long contexts such as
78- leveraging chunk overlap and/or a specialized stitching model.
79-
80- `chunk` breaks the input up into smaller chunks that fit within the model's
81- context window and then intelligently merges the results into a single
82- prediction at the cost of a minor accuracy drop.
74+ `auto` currently behaves the same as `drop_end`, dropping excess tokens from the
75+ end of input. In the future, `auto` may implement more sophisticated strategies
76+ such as chunking and context-aware stitching.
8377
8478 `drop_end` drops tokens from the end of input exceeding the model's maximum
8579 input length.
@@ -137,7 +131,7 @@ async def create(
137131 * ,
138132 model : Literal ["kanon-2-enricher" ],
139133 texts : Union [SequenceNotStr [str ], str ],
140- overflow_strategy : Optional [Literal ["auto" , "drop_end" , "chunk" ]] | Omit = omit ,
134+ overflow_strategy : Optional [Literal ["auto" , "drop_end" ]] | Omit = omit ,
141135 # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
142136 # The extra values given here take precedence over values defined on the client or passed to this method.
143137 extra_headers : Headers | None = None ,
@@ -159,15 +153,9 @@ async def create(
159153
160154 overflow_strategy: The strategy for handling content exceeding the model's maximum input length.
161155
162- `auto`, which is the recommended setting, currently behaves the same as `chunk`,
163- which intelligently breaks the input up into smaller chunks and then stitches
164- the results back together into a single prediction. In the future `auto` may
165- implement even more sophisticated strategies for handling long contexts such as
166- leveraging chunk overlap and/or a specialized stitching model.
167-
168- `chunk` breaks the input up into smaller chunks that fit within the model's
169- context window and then intelligently merges the results into a single
170- prediction at the cost of a minor accuracy drop.
156+ `auto` currently behaves the same as `drop_end`, dropping excess tokens from the
157+ end of input. In the future, `auto` may implement more sophisticated strategies
158+ such as chunking and context-aware stitching.
171159
172160 `drop_end` drops tokens from the end of input exceeding the model's maximum
173161 input length.
0 commit comments