oa.base

Base oa functionality

Functions

chat([prompt, model, messages, audio, ...])

complete(prompt[, model, best_of, echo, ...])

compute_price(model[, num_input_tokens, ...])

Compute the price of a model given the number of input and output tokens

dalle(prompt[, n, size, background, model, ...])

embeddings([texts, batch_size, egress, ...])

Get embeddings for a text or texts.

list_engine_ids([pattern])

List the available engine IDs.

model_information(model, information)

Get information about a model

normalize_text_input(texts)

Ensures the type of texts is an iterable of strings

text_is_valid(texts[, token_count, model, ...])

Check if text (a string or iterable of strings) is/are valid for a given model.

validate_texts_for_embeddings(texts, ...[, ...])

oa.base.compute_price(model, num_input_tokens=None, num_output_tokens=None)[source]

Compute the price of a model given the number of input and output tokens

oa.base.embeddings(texts=None, *, batch_size=2048, egress=None, batch_callback=None, validate=True, valid_text_getter=<function _raise_if_any_invalid>, model='text-embedding-3-small', client=None, dimensions=NOT_GIVEN, encoding_format=<openai.Omit object>, user=<openai.Omit object>, extra_headers=None, extra_query=None, extra_body=None, timeout=NOT_GIVEN)[source]

Get embeddings for a text or texts.

Parameters:
  • texts (Union[str, Iterable[str], Mapping[TypeVar(KT), str]]) – A string, an iterable of strings, or a dictionary of strings

  • egress (str | None) – A function that takes the embeddings and returns the desired output. If None, the output will be a list of embeddings. If False, no output will be returned, so the batch_callback had better be set to accumulate the results.

  • batch_callback (Callable[[int, list[list]], Any] | None) – A function that is called after each batch of embeddings is computed. This can be used for logging, saving, etc. One common use case is to save the intermediate results, in files, database, or in a list. This can be useful if you’re worried about the process failing and want to be able to resume from where you left off instead of having to start over (wasting time, and money). To accumulate the results in a list, you can set results = [] and then use a lambda function like this: batch_callback = lambda i, batch: results.extend(batch).

  • validate (bool | Callable | None) – If True, validate the text(s) before getting embeddings

  • valid_text_getter – A function that gets valid texts from the input texts

  • model – The model to use for embeddings

  • client – The OpenAI client to use

  • dimensions (int | None) – If given will reduce the dimensions of the full size embedding vectors to that size

  • batch_size (int | None) – The maximum number of texts to send in a single request

  • extra_embeddings_params – Extra parameters to pass to the embeddings API

>>> from functools import partial
>>> dimensions = 3
>>> embeddings_ = partial(embeddings, dimensions=dimensions, validate=True)

Test with a single word:

>>> text = "vector"
>>> result = embeddings_(text)
>>> result
[-0.4096039831638336, 0.3794299364089966, -0.8296127915382385]
>>> isinstance(result, list)
True
>>> len(result) == dimensions == 3
True

Test with a list of words

>>> texts = ["semantic", "vector"]
>>> result = embeddings_(texts)
>>> isinstance(result, list)
True
>>> len(result)
2

Two vectors; one for each word. Note that the second vector is the vector of “vector”, which we’ve seen before.

>>> result[1]
[-0.4096039831638336, 0.3794299364089966, -0.8296127915382385]
>>> len(result[1]) == dimensions == 3
True

Test with a dictionary of words

>>> texts = {"adj": "semantic", "noun": "vector"}
>>> result = embeddings_(texts)
>>> isinstance(result, dict)
True
>>> len(result)
2
>>> result["noun"]
[-0.4096039831638336, 0.3794299364089966, -0.8296127915382385]
>>> len(result["adj"]) == len(result["noun"]) == dimensions == 3
True

If you don’t specify texts, you will get a “partial” function that you can use later to compute embeddings for texts. This is useful if you want to set some parameters (like dimensions, validate, etc.) and then use the resulting function to compute embeddings for different texts later on.

>>> embeddings_with_10_dimensions = embeddings_(texts=None, dimensions=10)
>>> isinstance(embeddings_with_10_dimensions, Callable)
True
oa.base.list_engine_ids(pattern=None)[source]

List the available engine IDs. Optionally filter by a regex pattern.

oa.base.model_information(model, information)[source]

Get information about a model

oa.base.normalize_text_input(texts)[source]

Ensures the type of texts is an iterable of strings

Return type:

Iterable[str]

oa.base.text_is_valid(texts, token_count=True, *, model='text-embedding-3-small', max_tokens=None)[source]

Check if text (a string or iterable of strings) is/are valid for a given model.

Text is valid if

  • it is not empty

  • the number of tokens in the text is less than or equal to the max_tokens

Parameters:
  • texts (Union[str, Iterable[str], Mapping[TypeVar(KT), str]]) – a string or an iterable of strings

  • token_count – Specification of the token count of the text or texts

  • model (str) – The model to use for token count

  • max_tokens – The maximum number of tokens allowed by the model

If token_count is an integer, it will check if it is less than or equal to the max_tokens.

If token_count is True, it will compute the number of tokens in the text using the model specified by model and check if it is less than or equal to max_tokens.

If token_count is False, it will not check the token count.

If token_count is an iterable, it will apply the same mechanism as above, to each text in texts and the corresponding token count in token_count. This means both texts and token_count(s) must be of the same length.

Examples

>>> text_is_valid('Hello, world!')
True
>>> text_is_valid('')
False
>>> text_is_valid('Alice ' * 9000)
False
>>> text_is_valid('Alice ' * 9000, token_count=False)
True
>>> list(text_is_valid(['Bob', '', 'Alice ' * 9000]))
[True, False, False]
>>> list(text_is_valid(['Bob', '', 'Alice ' * 9000], token_count=False))
[True, False, True]