main
1# File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
2
3from __future__ import annotations
4
5from typing import Dict, Union, Iterable, Optional
6from typing_extensions import Literal, Required, TypedDict
7
8from .._types import SequenceNotStr
9from .chat.chat_completion_stream_options_param import ChatCompletionStreamOptionsParam
10
11__all__ = ["CompletionCreateParamsBase", "CompletionCreateParamsNonStreaming", "CompletionCreateParamsStreaming"]
12
13
14class CompletionCreateParamsBase(TypedDict, total=False):
15 model: Required[Union[str, Literal["gpt-3.5-turbo-instruct", "davinci-002", "babbage-002"]]]
16 """ID of the model to use.
17
18 You can use the
19 [List models](https://platform.openai.com/docs/api-reference/models/list) API to
20 see all of your available models, or see our
21 [Model overview](https://platform.openai.com/docs/models) for descriptions of
22 them.
23 """
24
25 prompt: Required[Union[str, SequenceNotStr[str], Iterable[int], Iterable[Iterable[int]], None]]
26 """
27 The prompt(s) to generate completions for, encoded as a string, array of
28 strings, array of tokens, or array of token arrays.
29
30 Note that <|endoftext|> is the document separator that the model sees during
31 training, so if a prompt is not specified the model will generate as if from the
32 beginning of a new document.
33 """
34
35 best_of: Optional[int]
36 """
37 Generates `best_of` completions server-side and returns the "best" (the one with
38 the highest log probability per token). Results cannot be streamed.
39
40 When used with `n`, `best_of` controls the number of candidate completions and
41 `n` specifies how many to return – `best_of` must be greater than `n`.
42
43 **Note:** Because this parameter generates many completions, it can quickly
44 consume your token quota. Use carefully and ensure that you have reasonable
45 settings for `max_tokens` and `stop`.
46 """
47
48 echo: Optional[bool]
49 """Echo back the prompt in addition to the completion"""
50
51 frequency_penalty: Optional[float]
52 """Number between -2.0 and 2.0.
53
54 Positive values penalize new tokens based on their existing frequency in the
55 text so far, decreasing the model's likelihood to repeat the same line verbatim.
56
57 [See more information about frequency and presence penalties.](https://platform.openai.com/docs/guides/text-generation)
58 """
59
60 logit_bias: Optional[Dict[str, int]]
61 """Modify the likelihood of specified tokens appearing in the completion.
62
63 Accepts a JSON object that maps tokens (specified by their token ID in the GPT
64 tokenizer) to an associated bias value from -100 to 100. You can use this
65 [tokenizer tool](/tokenizer?view=bpe) to convert text to token IDs.
66 Mathematically, the bias is added to the logits generated by the model prior to
67 sampling. The exact effect will vary per model, but values between -1 and 1
68 should decrease or increase likelihood of selection; values like -100 or 100
69 should result in a ban or exclusive selection of the relevant token.
70
71 As an example, you can pass `{"50256": -100}` to prevent the <|endoftext|> token
72 from being generated.
73 """
74
75 logprobs: Optional[int]
76 """
77 Include the log probabilities on the `logprobs` most likely output tokens, as
78 well the chosen tokens. For example, if `logprobs` is 5, the API will return a
79 list of the 5 most likely tokens. The API will always return the `logprob` of
80 the sampled token, so there may be up to `logprobs+1` elements in the response.
81
82 The maximum value for `logprobs` is 5.
83 """
84
85 max_tokens: Optional[int]
86 """
87 The maximum number of [tokens](/tokenizer) that can be generated in the
88 completion.
89
90 The token count of your prompt plus `max_tokens` cannot exceed the model's
91 context length.
92 [Example Python code](https://cookbook.openai.com/examples/how_to_count_tokens_with_tiktoken)
93 for counting tokens.
94 """
95
96 n: Optional[int]
97 """How many completions to generate for each prompt.
98
99 **Note:** Because this parameter generates many completions, it can quickly
100 consume your token quota. Use carefully and ensure that you have reasonable
101 settings for `max_tokens` and `stop`.
102 """
103
104 presence_penalty: Optional[float]
105 """Number between -2.0 and 2.0.
106
107 Positive values penalize new tokens based on whether they appear in the text so
108 far, increasing the model's likelihood to talk about new topics.
109
110 [See more information about frequency and presence penalties.](https://platform.openai.com/docs/guides/text-generation)
111 """
112
113 seed: Optional[int]
114 """
115 If specified, our system will make a best effort to sample deterministically,
116 such that repeated requests with the same `seed` and parameters should return
117 the same result.
118
119 Determinism is not guaranteed, and you should refer to the `system_fingerprint`
120 response parameter to monitor changes in the backend.
121 """
122
123 stop: Union[Optional[str], SequenceNotStr[str], None]
124 """Not supported with latest reasoning models `o3` and `o4-mini`.
125
126 Up to 4 sequences where the API will stop generating further tokens. The
127 returned text will not contain the stop sequence.
128 """
129
130 stream_options: Optional[ChatCompletionStreamOptionsParam]
131 """Options for streaming response. Only set this when you set `stream: true`."""
132
133 suffix: Optional[str]
134 """The suffix that comes after a completion of inserted text.
135
136 This parameter is only supported for `gpt-3.5-turbo-instruct`.
137 """
138
139 temperature: Optional[float]
140 """What sampling temperature to use, between 0 and 2.
141
142 Higher values like 0.8 will make the output more random, while lower values like
143 0.2 will make it more focused and deterministic.
144
145 We generally recommend altering this or `top_p` but not both.
146 """
147
148 top_p: Optional[float]
149 """
150 An alternative to sampling with temperature, called nucleus sampling, where the
151 model considers the results of the tokens with top_p probability mass. So 0.1
152 means only the tokens comprising the top 10% probability mass are considered.
153
154 We generally recommend altering this or `temperature` but not both.
155 """
156
157 user: str
158 """
159 A unique identifier representing your end-user, which can help OpenAI to monitor
160 and detect abuse.
161 [Learn more](https://platform.openai.com/docs/guides/safety-best-practices#end-user-ids).
162 """
163
164
165class CompletionCreateParamsNonStreaming(CompletionCreateParamsBase, total=False):
166 stream: Optional[Literal[False]]
167 """Whether to stream back partial progress.
168
169 If set, tokens will be sent as data-only
170 [server-sent events](https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events/Using_server-sent_events#Event_stream_format)
171 as they become available, with the stream terminated by a `data: [DONE]`
172 message.
173 [Example Python code](https://cookbook.openai.com/examples/how_to_stream_completions).
174 """
175
176
177class CompletionCreateParamsStreaming(CompletionCreateParamsBase):
178 stream: Required[Literal[True]]
179 """Whether to stream back partial progress.
180
181 If set, tokens will be sent as data-only
182 [server-sent events](https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events/Using_server-sent_events#Event_stream_format)
183 as they become available, with the stream terminated by a `data: [DONE]`
184 message.
185 [Example Python code](https://cookbook.openai.com/examples/how_to_stream_completions).
186 """
187
188
189CompletionCreateParams = Union[CompletionCreateParamsNonStreaming, CompletionCreateParamsStreaming]