Generate a response for a given prompt

Usage

generate(
  model,
  prompt,
  suffix = "",
  images = "",
  system = "",
  template = "",
  context = list(),
  stream = FALSE,
  raw = FALSE,
  keep_alive = "5m",
  output = c("resp", "jsonlist", "raw", "df", "text", "req"),
  endpoint = "/api/generate",
  host = NULL,
  ...
)

Arguments

model: A character string of the model name such as "llama3".
prompt: A character string of the prompt like "The sky is..."
suffix: A character string after the model response. Default is "".
images: A path to an image file to include in the prompt. Default is "".
system: A character string of the system prompt (overrides what is defined in the Modelfile). Default is "".
template: A character string of the prompt template (overrides what is defined in the Modelfile). Default is "".
context: A list of context from a previous response to include previous conversation in the prompt. Default is an empty list.
stream: Enable response streaming. Default is FALSE.
raw: If TRUE, no formatting will be applied to the prompt. You may choose to use the raw parameter if you are specifying a full templated prompt in your request to the API. Default is FALSE.
keep_alive: The time to keep the connection alive. Default is "5m" (5 minutes).
output: A character vector of the output format. Default is "resp". Options are "resp", "jsonlist", "raw", "df", "text", "req" (httr2_request object).
endpoint: The endpoint to generate the completion. Default is "/api/generate".
host: The base URL to use. Default is NULL, which uses Ollama's default base URL.
...: Additional options to pass to the model.

Value

A response in the format specified in the output parameter.

References

API documentation

Examples

if (FALSE) { # test_connection()$status_code == 200
# text prompt
generate("llama3", "The sky is...", stream = FALSE, output = "df")
# stream and increase temperature
generate("llama3", "The sky is...", stream = TRUE, output = "text", temperature = 2.0)

# image prompt
# something like "image1.png"
image_path <- file.path(system.file("extdata", package = "ollamar"), "image1.png")
# use vision or multimodal model such as https://ollama.com/benzie/llava-phi-3
generate("benzie/llava-phi-3:latest", "What is in the image?", images = image_path, output = "text")
}