https://github.com/dfalbel/gptneox

What the Package Does (One Line, Title Case)
https://github.com/dfalbel/gptneox

Last synced: 23 days ago
JSON representation

What the Package Does (One Line, Title Case)

• Host: GitHub
• URL: https://github.com/dfalbel/gptneox
• Owner: dfalbel
• Created: 2023-05-01T13:29:32.000Z (over 1 year ago)
• Default Branch: master
• Last Pushed: 2023-05-11T07:49:42.000Z (over 1 year ago)
• Last Synced: 2024-10-05T04:42:07.743Z (about 1 month ago)
• Language: R
• Size: 34.2 KB
• Stars: 8
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.Rmd

Awesome Lists containing this project

• awesome-ChatGPT-repositories - gptneox - What the Package Does (One Line, Title Case) (Others)

README

        
---

output: github_document

---

```{r, include = FALSE}

knitr::opts_chunk$set(

  collapse = TRUE,

  comment = "#>",

  fig.path = "man/figures/README-",

  out.width = "100%"

)

```

# gptneox

gptneox is an R torch implementation of GPTNeoX. It follows closely the [implementation](https://huggingface.co/docs/transformers/model_doc/gpt_neox)

available in HuggingFace and can be used to load pre-trained models defined there.

## Installation

You can install the development version of gptneox like so:

``` r

remotes::install_github("dfalbel/gptneox")

```

Note: This package requires [`tok`](https://github.com/dfalbel/tok), which in turns requires a Rust installation to

be installed. Follow instruction in the [`tok` repository](https://github.com/dfalbel/tok) to get it installed.

## Example

Here's an example using the StabilityAI 3B parameters model. Note that this model

is not tuned to provide chat like completions, thus you should write prompts that look

more like autocomplete queries. You can load other GPTNeoX models, including those

that have been trained for chat like completions.

```{r example}

library(gptneox)

torch::torch_manual_seed(1)

repo <- "stabilityai/stablelm-base-alpha-3b"

model <- gpt_neox_from_pretrained(repo)

tok <- gpt_neox_tokenizer_from_pretrained(repo)

gen <- gpt_neox_generate(

  model, tok, 

  "The R language was created", 

  config = list(max_new_tokens = 100)

)

```