How to use it with Ollama?

[diecasfranco]

Hi all, by default, Ask is using OpenAI. How to use it with Ollama?

[moodymudskipper]

Thanks for your question, I should put this in the README, let's keep this issue open until then. First you should make sure the model you want to use is installed: install the ollama app ( https://ollama.com/download ) then pull the model you need from the terminal, for instance ollama pull llama3.1 (this needs to be done once only) All ask*() functions have a model argument that you can set for instance to "llama3.1" or "llama3.2". I believe I only tested it with llama3.1. Rather than providing the model each time you might prefer to set it through options, for instance options(ask.model = "llama3.1").

Please tell me if these instructions work for you.

[MislavSag]

This works for me. Is it possible to include arbitrary files in context? For example pdf files.

[diecasfranco]

Now it is working. Thank you

[moodymudskipper]

There is no context_pdf() yet in the package, but I will add it, this seems to work (only tested with example below, which you should be able to reproduce):

# ... forwarded to pdftools::pdf_text(), pdftools::pdf_ocr_text(), or/and pdftools::pdf_ocr_data()
context_pdf <- function(
    file, 
    pages = NULL,
    opw = "",
    upw = "",
    dpi = 600,
    language = "eng",
    options = NULL,
    type = "text") {
  rlang::check_installed("pdftools")
  type = rlang::arg_match(type, c("text", "ocr_text", "ocr_data"), multiple = TRUE)
  contexts <- list()
  for (i in seq_along(type)) {
    contexts[[i]] <-   switch(
      type[[i]],
      text = context('Pdf file OCRed text content: {file}' := pdftools::pdf_text(
        pdf = file, 
        opw = opw, 
        upw = upw
      )),
      ocr_text = context('Pdf file OCRed text content: {file}' := pdftools::pdf_ocr_text(
        pdf = file, 
        opw = opw, 
        upw = upw,
        dpi = dpi,
        language = language,
        options = options
      )),
      ocr_data = context('Pdf file OCRed data content: {file}' := pdftools::pdf_ocr_data(
        pdf = file, 
        opw = opw, 
        upw = upw,
        dpi = dpi,
        language = language,
        options = options
      )),
    )
  }
  context(!!!contexts)
}

ask("what is this about?", context_pdf("http://arxiv.org/pdf/1403.2805.pdf"))

To feed images from the pdf to the model we need something else we need a bit more work but it is possible too.

[moodymudskipper]

btw @MislavSag you mentioned arbitrary files, if you need support for other file types, please open new tickets, these are easy to implement so I should be able to ship them quite fast.

[MislavSag]

Thanks @moodymudskipper . I have tried your function and it works. The answers are pretty bad, but this due to pdf parsing I suppose.