Run Llama.cpp inside neovim

Llama.lua

local function get_visual_selection()
  local s_start = vim.fn.getpos("'<")
  local s_end = vim.fn.getpos("'>")
  local n_lines = math.abs(s_end[2] - s_start[2]) + 1
  local lines = vim.api.nvim_buf_get_lines(0, s_start[2] - 1, s_end[2], false)
  lines[1] = string.sub(lines[1], s_start[3], -1)
  if n_lines == 1 then
    lines[n_lines] = string.sub(lines[n_lines], 1, s_end[3] - s_start[3] + 1)
  else
    lines[n_lines] = string.sub(lines[n_lines], 1, s_end[3])
  end
  return table.concat(lines, '\n')
end


vim.api.nvim_create_user_command('Llama',
  function(opts)
    local prompt = vim.fn.input "prompt: "
    local select = get_visual_selection()
    local cmd = '~/Projects/agi/llama.cpp/llama-cli -m ~/Public/models/gemma_2_chinese_chat_q4_k_m.gguf --log-disable --temp 0.5 -p "' .. prompt .. select .. '"'
    vim.api.nvim_command('term ' .. cmd)
  end,
  { range = true }
)