blob: adbe24c1aa6be460c329084f7651e45686483063 [file] [log] [blame]
ben-aaron18885c32a02022-09-10 20:30:30 +02001#' Makes a single completion request to the GPT-3 API
ben-aaron1883818e7c2022-09-08 17:49:01 +02002#'
3#' @description
ben-aaron1888f5f2002022-09-10 21:26:23 +02004#' `gpt3_simple_request()` sends a single [completion request](https://beta.openai.com/docs/api-reference/completions) to the Open AI GPT-3 API.
ben-aaron18885c32a02022-09-10 20:30:30 +02005#' @details For a general guide on the completion requests, see [https://beta.openai.com/docs/guides/completion](https://beta.openai.com/docs/guides/completion). This function provides you with an R wrapper to send requests with the full range of request parameters as detailed on [https://beta.openai.com/docs/api-reference/completions](https://beta.openai.com/docs/api-reference/completions) and reproduced below.
6#'
7#' For the `best_of` parameter: When used with n, best_of controls the number of candidate completions and n specifies how many to return – best_of must be greater than n. Note that this is handled by the wrapper automatically if(best_of <= n){ best_of = n}.
8#'
9#' Parameters not included/supported:
10#' - `logit_bias`: [https://beta.openai.com/docs/api-reference/completions/create#completions/create-logit_bias](https://beta.openai.com/docs/api-reference/completions/create#completions/create-logit_bias)
11#' - `echo`: [https://beta.openai.com/docs/api-reference/completions/create#completions/create-echo](https://beta.openai.com/docs/api-reference/completions/create#completions/create-echo)
12#' - `stream`: [https://beta.openai.com/docs/api-reference/completions/create#completions/create-stream](https://beta.openai.com/docs/api-reference/completions/create#completions/create-stream)
13#'
14#' @param prompt_input character that contains the prompt to the GPT-3 request
15#' @param model a character vector that indicates the [model](https://beta.openai.com/docs/models/gpt-3) to use; one of "text-davinci-002" (default), "text-curie-001", "text-babbage-001" or "text-ada-001"
16#' @param output_type character determining the output provided: "complete" (default), "text" or "meta"
17#' @param suffix character (default: NULL) (from the official API documentation: _The suffix that comes after a completion of inserted text_)
18#' @param max_tokens numeric (default: 100) indicating the maximum number of tokens that the completion request should return (from the official API documentation: _The maximum number of tokens to generate in the completion. The token count of your prompt plus max_tokens cannot exceed the model's context length. Most models have a context length of 2048 tokens (except for the newest models, which support 4096)_)
19#' @param temperature numeric (default: 0.9) specifying the sampling strategy of the possible completions (from the official API documentation: _What sampling temperature to use. Higher values means the model will take more risks. Try 0.9 for more creative applications, and 0 (argmax sampling) for ones with a well-defined answer. We generally recommend altering this or top_p but not both._)
20#' @param top_p numeric (default: 1) specifying sampling strategy as an alternative to the temperature sampling (from the official API documentation: _An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. We generally recommend altering this or temperature but not both._)
21#' @param n numeric (default: 1) specifying the number of completions per request (from the official API documentation: _How many completions to generate for each prompt. **Note: Because this parameter generates many completions, it can quickly consume your token quota.** Use carefully and ensure that you have reasonable settings for max_tokens and stop._)
22#' @param logprobs numeric (default: NULL) (from the official API documentation: _Include the log probabilities on the logprobs most likely tokens, as well the chosen tokens. For example, if logprobs is 5, the API will return a list of the 5 most likely tokens. The API will always return the logprob of the sampled token, so there may be up to logprobs+1 elements in the response. The maximum value for logprobs is 5. If you need more than this, please contact support@openai.com and describe your use case._)
23#' @param stop character or character vector (default: NULL) that specifies after which character value when the completion should end (from the official API documentation: _Up to 4 sequences where the API will stop generating further tokens. The returned text will not contain the stop sequence._)
24#' @param presence_penalty numeric (default: 0) between -2.00 and +2.00 to determine the penalisation of repetitiveness if a token already exists (from the official API documentation: _Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics._). See also: [https://beta.openai.com/docs/api-reference/parameter-details](https://beta.openai.com/docs/api-reference/parameter-details)
25#' @param frequency_penalty numeric (default: 0) between -2.00 and +2.00 to determine the penalisation of repetitiveness based on the frequency of a token in the text already (from the official API documentation: _Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim._). See also: [https://beta.openai.com/docs/api-reference/parameter-details](https://beta.openai.com/docs/api-reference/parameter-details)
26#' @param best_of numeric (default: 1) that determines the space of possibilities from which to select the completion with the highest probability (from the official API documentation: _Generates `best_of` completions server-side and returns the "best" (the one with the highest log probability per token)_). See details.
27#'
28#' @return A list with two data tables (if `output_type` is the default "complete"): [[1]] contains the data table with the columns `n` (= the mo. of `n` responses requested), `prompt` (= the prompt that was sent), and `gpt3` (= the completion as returned from the GPT-3 model). [[2]] contains the meta information of the request, including the request id, the parameters of the request and the token usage of the prompt (`tok_usage_prompt`), the completion (`tok_usage_completion`) and the total usage (`tok_usage_total`).
29#'
30#' If `output_type` is "text", only the data table in slot [[1]] is returned.
31#'
32#' If `output_type` is "meta", only the data table in slot [[2]] is returned.
ben-aaron1883818e7c2022-09-08 17:49:01 +020033#' @examples
ben-aaron18885c32a02022-09-10 20:30:30 +020034#' # First authenticate with your API key via `gpt3_authenticate('pathtokey')`
35#'
36#' # Once authenticated:
37#'
38#' ## Simple request with defaults:
39#' gpt3_simple_request(prompt_input = 'How old are you?')
40#'
41#' ## Instruct GPT-3 to write ten research ideas of max. 150 tokens with some controls:
42#'gpt3_simple_request(prompt_input = 'Write a research idea about using text data to understand human behaviour:', temperature = 0.8, n = 10, max_tokens = 150)
43#'
44#' ## For fully reproducible results, we need `temperature = 0`, e.g.:
45#' gpt3_simple_request(prompt_input = 'Finish this sentence:/n There is no easier way to learn R than', temperature = 0.0, max_tokens = 50)
46#'
47#' ## The same example with a different GPT-3 model:
48#' gpt3_simple_request(prompt_input = 'Finish this sentence:/n There is no easier way to learn R than', model = 'text-babbage-001', temperature = 0.0, max_tokens = 50)
ben-aaron1883818e7c2022-09-08 17:49:01 +020049#' @export
ben-aaron18885c32a02022-09-10 20:30:30 +020050gpt3_simple_request = function(prompt_input
51 , model = 'text-davinci-002'
52 , output_type = 'complete'
53 , suffix = NULL
54 , max_tokens = 100
55 , temperature = 0.9
56 , top_p = 1
57 , n = 1
58 , logprobs = NULL
59 , stop = NULL
60 , presence_penalty = 0
61 , frequency_penalty = 0
62 , best_of = 1){
ben-aaron1883818e7c2022-09-08 17:49:01 +020063
ben-aaron18885c32a02022-09-10 20:30:30 +020064 #check for request issues with `n` and `best_of`
65 if(best_of < n){
66 best_of = n
67 message('To avoid an `invalid_request_error`, `best_of` was set to equal `n`')
68 }
69
70 if(temperature == 0 & n > 1){
71 n = 1
72 message('You are running the deterministic model, so `n` was set to 1 to avoid unnecessary token quota usage.')
73 }
74
75 parameter_list = list(prompt = prompt_input
76 , model = model
77 , suffix = suffix
78 , max_tokens = max_tokens
79 , temperature = temperature
80 , top_p = top_p
81 , n = n
82 , logprobs = logprobs
83 , stop = stop
84 , presence_penalty = presence_penalty
85 , frequency_penalty = frequency_penalty
86 , best_of = best_of)
ben-aaron1883818e7c2022-09-08 17:49:01 +020087
88 request_base = httr::POST(url = url.completions
ben-aaron18885c32a02022-09-10 20:30:30 +020089 , body = parameter_list
90 , httr::add_headers(Authorization = paste("Bearer", api_key))
91 , encode = "json")
92
93 request_content = httr::content(request_base)
94
95 if(n == 1){
96 core_output = data.table::data.table('n' = 1
97 , 'prompt' = prompt_input
98 , 'gpt3' = request_content$choices[[1]]$text)
99 } else if(n > 1){
100
101 core_output = data.table::data.table('n' = 1:n
102 , 'prompt' = rep(prompt_input, n)
103 , 'gpt3' = rep("", n))
104
105 for(i in 1:n){
106 core_output$gpt3[i] = request_content$choices[[i]]$text
107 }
108
109 }
ben-aaron1883818e7c2022-09-08 17:49:01 +0200110
111
ben-aaron18885c32a02022-09-10 20:30:30 +0200112 meta_output = data.table::data.table('request_id' = request_content$id
113 , 'object' = request_content$object
114 , 'model' = request_content$model
115 , 'param_prompt' = prompt_input
116 , 'param_model' = model
117 , 'param_suffix' = suffix
118 , 'param_max_tokens' = max_tokens
119 , 'param_temperature' = temperature
120 , 'param_top_p' = top_p
121 , 'param_n' = n
122 , 'param_logprobs' = logprobs
123 , 'param_stop' = stop
124 , 'param_presence_penalty' = presence_penalty
125 , 'param_frequency_penalty' = frequency_penalty
126 , 'param_best_of' = best_of
127 , 'tok_usage_prompt' = request_content$usage$prompt_tokens
128 , 'tok_usage_completion' = request_content$usage$completion_tokens
129 , 'tok_usage_total' = request_content$usage$total_tokens)
130
131 if(output_type == 'complete'){
132 output = list(core_output
133 , meta_output)
134 } else if(output_type == 'meta'){
135 output = meta_output
136 } else if(output_type == 'text'){
137 output = core_output
ben-aaron1883818e7c2022-09-08 17:49:01 +0200138 }
139
140 return(output)
141
142}