A Neovim plugin that provides local LLM-powered code completion. This is a Lua rewrite of ggml-org/llama.vim.
- Neovim 0.9+
- A running local LLM server (e.g., llama.cpp) check here
- LRU caching of requests
- Treesitter based extra context
- Play nicely with blink.cmp and other plugins
- Profiling
{
'georg3tom/llama.nvim',
config = function()
require('llama').setup({
-- Your configuration options here
endpoint = 'http://localhost:8012/infill',
auto_fim = true,
n_predict = 128
})
end
}use {
'georg3tom/llama.nvim',
config = function()
require('llama').setup({
-- Your configuration options here
})
end
}The plugin can be configured with the following options:
require('llama').setup({
endpoint = '127.0.0.1:8012/infill', -- LLM server endpoint
api_key = '', -- Optional API key
n_prefix = 256, -- Lines of context before cursor
n_suffix = 64, -- Lines of context after cursor
n_predict = 128, -- Maximum tokens to predict
stop_strings = {}, -- Strings that stop generation when encountered
auto_fim = true, -- Auto-trigger completion
max_cache_keys = 250, -- Size of the cache
keymap_trigger = '<C-F>', -- Trigger completion keymap
keymap_accept_full = '<Tab>', -- Accept full completion
keymap_accept_line = '<S-Tab>', -- Accept line completion
keymap_accept_word = '<C-B>' -- Accept word completion
})To force completions to be single-line only, configure the stop_strings option to include a newline character:
require('llama').setup({
-- Other configuration options
stop_strings = { "\n" },
})<C-F>: Trigger completion manually<Tab>: Accept the full completion<S-Tab>: Accept line completion<C-B>: Accept word completion
:LlamaEnable: Enable the plugin:LlamaDisable: Disable the plugin:LlamaToggle: Toggle the plugin:LlamaInfo: Show logs