llama.vim

" vim: ts=4 sts=4 expandtab
" colors (adjust to your liking)

" fim colors
highlight default llama_hl_fim_hint guifg=#ff772f ctermfg=202
highlight default llama_hl_fim_info guifg=#77ff2f ctermfg=119

 " instruct colors for selected block
 highlight default llama_hl_inst_src guibg=#554433 ctermbg=236

 " virtual text colors for instructions
 highlight default llama_hl_inst_virt_proc  guifg=#77ff2f ctermfg=119
 highlight default llama_hl_inst_virt_gen   guifg=#77ff2f ctermfg=119
 highlight default llama_hl_inst_virt_ready guifg=#ff772f ctermfg=202

" general parameters:
"
"   endpoint_fim:     llama.cpp server endpoint for FIM completion
"   endpoint_inst:    llama.cpp server endpoint for instruction completion
"   model_fim:        model name in case when multiple models are loaded (optional, recommended: Qwen3 Coder 30B)
"   model_inst:       instruction model name (optional, recommended: gpt-oss-120b)
"   api_key:          llama.cpp server api key (optional)
"   n_prefix:         number of lines before the cursor location to include in the local prefix
"   n_suffix:         number of lines after  the cursor location to include in the local suffix
"   n_predict:        max number of tokens to predict
"   stop_strings_fim  return the result immediately as soon as any of these strings are encountered in the generated text for FIM completions
"   stop_strings_inst return the result immediately as soon as any of these strings are encountered in the generated text for instruction completions
"   n_cmpl:           number of completions to cache per position (ring buffer, default: 1)
"   t_max_prompt_ms:  max alloted time for the prompt processing (TODO: not yet supported)
"   t_max_predict_ms: max alloted time for the prediction
"   show_info:        show extra info about the inference (0 - disabled, 1 - statusline, 2 - inline)
"   auto_fim:         trigger FIM completion automatically on cursor movement
"   max_line_suffix:  do not auto-trigger FIM completion if there are more than this number of characters to the right of the cursor
"   max_cache_keys:   max number of cached completions to keep in result_cache
"   enable_at_startup: enable llama.vim functionality at startup (default: v:true)
"
" ring buffer of chunks, accumulated with time upon:
"
"  - completion request
"  - yank
"  - entering a buffer
"  - leaving a buffer
"  - writing a file
"
" parameters for the ring-buffer with extra context:
"
"   ring_n_chunks:    max number of chunks to pass as extra context to the server (0 to disable)
"   ring_chunk_size:  max size of the chunks (in number of lines)
"                     note: adjust these numbers so that you don't overrun your context
"                           at ring_n_chunks = 64 and ring_chunk_size = 64 you need ~32k context
"   ring_scope:       the range around the cursor position (in number of lines) for gathering chunks after FIM
"   ring_update_ms:   how often to process queued chunks in normal mode
"
" keymaps parameters (empty string to disable):
"
"   keymap_fim_trigger:     keymap to trigger the completion, default: <C-F>
"   keymap_fim_accept_full: keymap to accept full suggestion, default: <Tab>
"   keymap_fim_accept_line: keymap to accept line suggestion, default: <S-Tab>
"   keymap_fim_accept_word: keymap to accept word suggestion, default: <C-B>
"   keymap_fim_next:        keymap to cycle to next completion,  default: <C-J>
"   keymap_fim_prev:        keymap to cycle to prev completion,  default: <C-K>
"   keymap_debug_toggle:    keymap to toggle the debug pane,  default: null
"   keymap_inst_trigger:    keymap to trigger the instruction command, default: <leader>lli
"   keymap_inst_rerun:      keymap to rerun the instruction, default: <leader>llr
"   keymap_inst_continue:   keymap to continue the instruction, default: <leader>llc
"   keymap_inst_accept:     keymap to accept the instruction, default: <Tab>
"   keymap_inst_cancel:     keymap to cancel the instruction, default: <Esc>
"
let s:default_config = {
    \ 'endpoint_fim':           'http://127.0.0.1:8012/infill',
    \ 'endpoint_inst':          'http://127.0.0.1:8012/v1/chat/completions',
    \ 'model_fim':              '',
    \ 'model_inst':             '',
    \ 'api_key':                '',
    \ 'n_prefix':               256,
    \ 'n_suffix':               64,
    \ 'n_predict':              128,
    \ 'stop_strings_fim':       [],
    \ 'stop_strings_inst':      [],
    \ 'n_cmpl':                 1,
    \ 't_max_prompt_ms':        500,
    \ 't_max_predict_ms':       1000,
    \ 'show_info':              2,
    \ 'auto_fim':               v:true,
    \ 'max_line_suffix':        8,
    \ 'max_cache_keys':         250,
    \ 'ring_n_chunks':          16,
    \ 'ring_chunk_size':        64,
    \ 'ring_scope':             1024,
    \ 'ring_update_ms':         1000,
    \ 'keymap_fim_trigger':     "<leader>llf",
    \ 'keymap_fim_accept_full': "<Tab>",
    \ 'keymap_fim_accept_line': "<S-Tab>",
    \ 'keymap_fim_accept_word': "<leader>ll]",
    \ 'keymap_fim_next':        "<C-J>",
    \ 'keymap_fim_prev':        "<C-K>",
    \ 'keymap_inst_trigger':    "<leader>lli",
    \ 'keymap_inst_rerun':      "<leader>llr",
    \ 'keymap_inst_continue':   "<leader>llc",
    \ 'keymap_inst_accept':     "<Tab>",
    \ 'keymap_inst_cancel':     "<Esc>",
    \ 'keymap_debug_toggle':    "<leader>lld",
    \ 'enable_at_startup':      v:true,
    \ }

let llama_config = get(g:, 'llama_config', s:default_config)

" rename deprecated keys in `llama_config`.
let s:renames = {
      \ 'endpoint'           : 'endpoint_fim',
      \ 'model'              : 'model_fim',
      \ 'keymap_trigger'     : 'keymap_fim_trigger',
      \ 'keymap_accept_full' : 'keymap_fim_accept_full',
      \ 'keymap_accept_line' : 'keymap_fim_accept_line',
      \ 'keymap_accept_word' : 'keymap_fim_accept_word',
      \ 'keymap_debug'       : 'keymap_debug_toggle',
      \ 'stop_strings'       : 'stop_strings_fim',
      \ }

for [old_key, new_key] in items(s:renames)
    if has_key(llama_config, old_key)
        let llama_config[new_key] = llama_config[old_key]

        call remove(llama_config, old_key)

        echohl WarningMsg
        echomsg printf(
            \ 'llama.vim: %s is deprecated, use %s instead',
            \ old_key, new_key)
        echohl None
    endif
endfor

let g:llama_config = extendnew(s:default_config, llama_config, 'force')

let s:llama_enabled = v:false

" containes cached responses from the server
" each key maps to a ring buffer of up to n_cmpl individual completion responses
" used to avoid re-computing the same completions and to also create new completions with similar context
" ref: https://github.com/ggml-org/llama.vim/pull/18
let g:cache_data = {}
let g:cache_lru_order = []

" insert a single completion response into the cache ring buffer for a key
" the ring buffer holds up to g:llama_config.n_cmpl entries per key
function! s:cache_insert(key, response)
    " Check if we need to evict a key
    if !has_key(g:cache_data, a:key) && len(keys(g:cache_data)) >= g:llama_config.max_cache_keys
        " Get the least recently used key (first in order list)
        let l:lru_key = g:cache_lru_order[0]
        " Remove from cache data
        call remove(g:cache_data, l:lru_key)
        " Remove from LRU order
        call remove(g:cache_lru_order, 0)
    endif

    " Initialize ring buffer for this key if needed
    if !has_key(g:cache_data, a:key)
        let g:cache_data[a:key] = []
    endif

    " skip if a completion with the same content already exists for this key
    let l:new_content = get(a:response, 'content', '')
    for l:existing in g:cache_data[a:key]
        if get(l:existing, 'content', '') ==# l:new_content
            return
        endif
    endfor

    " Add to ring buffer, evict oldest if full
    if len(g:cache_data[a:key]) >= g:llama_config.n_cmpl
        call remove(g:cache_data[a:key], 0)
    endif
    call add(g:cache_data[a:key], a:response)

    " Update LRU order - remove key if it exists and add to end (most recent)
    call filter(g:cache_lru_order, 'v:val !=# a:key')
    call add(g:cache_lru_order, a:key)
endfunction

" Get all cached completions for a key (ring buffer), or v:null if not found
function! s:cache_get(key)
    if !has_key(g:cache_data, a:key)
        return v:null
    endif

    " Update LRU order - remove key if it exists and add to end (most recent)
    call filter(g:cache_lru_order, 'v:val !=# a:key')
    call add(g:cache_lru_order, a:key)

    return g:cache_data[a:key]
endfunction

" Count total cached entries across all keys
function! s:cache_count()
    let l:total = 0
    for l:entries in values(g:cache_data)
        let l:total += len(l:entries)
    endfor
    return l:total
endfunction

" get the number of leading spaces of a string
function! s:get_indent(str)
    let l:count = 0
    for i in range(len(a:str))
        if a:str[i] == "\t"
            let l:count += &tabstop
        elseif a:str[i] == " "
            let l:count += 1
        else
            break
        endif
    endfor

    return l:count
endfunction

function! s:rand(i0, i1) abort
    return a:i0 + rand() % (a:i1 - a:i0 + 1)
endfunction

function! llama#disable()
    call llama#fim_hide()

    autocmd! llama

    " TODO: these unmaps don't seem to work properly
    if g:llama_config.keymap_fim_trigger != ''
        exe "silent! iunmap <buffer> " .. g:llama_config.keymap_fim_trigger
    endif
    if g:llama_config.keymap_fim_accept_full != ''
        exe "silent! iunmap <buffer> " .. g:llama_config.keymap_fim_accept_full
    endif
    if g:llama_config.keymap_fim_accept_line != ''
        exe "silent! iunmap <buffer> " .. g:llama_config.keymap_fim_accept_line
    endif
    if g:llama_config.keymap_fim_accept_word != ''
        exe "silent! iunmap <buffer> " .. g:llama_config.keymap_fim_accept_word
    endif
    if g:llama_config.keymap_fim_next != ''
        exe "silent! iunmap <buffer> " .. g:llama_config.keymap_fim_next
    endif
    if g:llama_config.keymap_fim_prev != ''
        exe "silent! iunmap <buffer> " .. g:llama_config.keymap_fim_prev
    endif

    if g:llama_config.keymap_debug_toggle != ''
        exe "silent!  unmap          " .. g:llama_config.keymap_debug_toggle
    endif
    if g:llama_config.keymap_inst_trigger != ''
        exe "silent! vunmap          " .. g:llama_config.keymap_inst_trigger
    endif
    if g:llama_config.keymap_inst_rerun != ''
        exe "silent!  unmap          " .. g:llama_config.keymap_inst_rerun
    endif
    if g:llama_config.keymap_inst_continue != ''
        exe "silent!  unmap          " .. g:llama_config.keymap_inst_continue
    endif
    if g:llama_config.keymap_inst_accept != ''
        exe "silent!  unmap          " .. g:llama_config.keymap_inst_accept
    endif
    if g:llama_config.keymap_inst_cancel != ''
        exe "silent!  unmap          " .. g:llama_config.keymap_inst_cancel
    endif

    let s:llama_enabled = v:false

    call llama#debug_log('plugin disabled')
endfunction

function! llama#toggle()
    if s:llama_enabled
        call llama#disable()
    else
        call llama#enable()
    endif
endfunction

function! llama#toggle_auto_fim()
    if !s:llama_enabled
        return
    endif

    let g:llama_config.auto_fim = !g:llama_config.auto_fim

    call llama#setup_autocmds()
endfunction

function! llama#status()
    let l:fim_base = substitute(g:llama_config.endpoint_fim, '/infill$', '', '')
    let l:inst_base = substitute(g:llama_config.endpoint_inst, '/v1/chat/completions$', '', '')

    let s:status_messages = {'fim': '', 'inst': '', 'count': 0}

    " If both endpoints are the same then we are running llama-server as a
    " router. Otherwise each are independent servers.
    if l:fim_base ==# l:inst_base
        let l:models_url = l:fim_base . '/v1/models'
        call s:fetch_status(l:models_url, 'both')
    else
        let l:fim_url = l:fim_base . '/v1/models'
        let l:inst_url = l:inst_base . '/v1/models'

        call s:fetch_status(l:fim_url, 'fim')
        call s:fetch_status(l:inst_url, 'inst')
    endif
endfunction

function! s:fetch_status(url, type)
    let l:curl_command = [
        \ "curl",
        \ "--silent",
        \ "--max-time", "3",
        \ "--request", "GET",
        \ "--url", a:url,
        \ ]

    if s:ghost_text_nvim
        let l:job = jobstart(l:curl_command, {
            \ 'on_stdout': function('s:status_on_response', [a:type]),
            \ 'on_exit': function('s:status_on_exit', [a:type]),
            \ 'stdout_buffered': v:true
            \ })
    elseif s:ghost_text_vim
        let l:job = job_start(l:curl_command, {
            \ 'out_cb': function('s:status_on_response', [a:type]),
            \ 'exit_cb': function('s:status_on_exit', [a:type])
            \ })
    endif
endfunction

function! s:status_on_response(type, job_id, data, event = v:null)
    if s:ghost_text_nvim
        let l:raw = join(a:data, "\n")
    elseif s:ghost_text_vim
        let l:raw = a:data
    endif

    if empty(l:raw)
        return
    endif

    try
        let l:response = json_decode(l:raw)
        let l:models = get(l:response, 'data', [])

        if a:type ==# 'both'
            let l:fim_status = s:get_model_status(g:llama_config.model_fim, l:models)
            let l:inst_status = s:get_model_status(g:llama_config.model_inst, l:models)
            echo 'FIM model (' . g:llama_config.model_fim . '): ' . l:fim_status . ', Instruction model (' . g:llama_config.model_inst . '): ' . l:inst_status
        elseif a:type ==# 'fim'
            let l:fim_status = s:get_model_status(g:llama_config.model_fim, l:models)
            let s:status_messages.fim = 'FIM model (' . g:llama_config.model_fim . '): ' . l:fim_status
            let s:status_messages.count += 1
            call s:display_status_if_ready()
        elseif a:type ==# 'inst'
            let l:inst_status = s:get_model_status(g:llama_config.model_inst, l:models)
            let s:status_messages.inst = 'Instruction model (' . g:llama_config.model_inst . '): ' . l:inst_status
            let s:status_messages.count += 1
            call s:display_status_if_ready()
        endif
    catch
        echom 'Error parsing status response: ' . v:exception
    endtry
endfunction

function! s:status_on_exit(type, job_id, exit_code, event = v:null)
    if a:exit_code != 0
        call llama#debug_log('status_on_exit (' . a:type . '): curl failed with exit code ' . a:exit_code)
        if a:type ==# 'both'
            echo 'LlamaStatus: ❌ Server not reachable'
        elseif a:type ==# 'fim'
            let s:status_messages.fim = 'FIM server: ❌ Not reachable'
            let s:status_messages.count += 1
            call s:display_status_if_ready()
        elseif a:type ==# 'inst'
            let s:status_messages.inst = 'Instruction server: ❌ Not reachable'
            let s:status_messages.count += 1
            call s:display_status_if_ready()
        endif
    endif
endfunction

function! s:display_status_if_ready()
    " Wait for both FIM and instruction responses.
    if s:status_messages.count >= 2
        echo s:status_messages.fim . ', ' . s:status_messages.inst
    endif
endfunction

function! s:get_model_status(model_name, models)
    if empty(a:model_name)
        return '❌ Not configured'
    endif

    for l:model in a:models
        if get(l:model, 'id', '') ==# a:model_name || index(get(l:model, 'tags', []), a:model_name) >= 0 || index(get(l:model, 'aliases', []), a:model_name) >= 0
            if has_key(l:model, 'status')
                let l:status_value = get(get(l:model, 'status', {}), 'value', 'unknown')
                return l:status_value ==# 'loaded' ? '✅ Ready' : l:status_value
            else
                return '✅ Ready'
            endif
        endif
    endfor

    if len(a:models) == 1
        let l:model = a:models[0]
        if has_key(l:model, 'status')
            let l:status_value = get(get(l:model, 'status', {}), 'value', 'unknown')
            return l:status_value ==# 'loaded' ? '✅ Ready' : l:status_value
        else
            return '✅ Ready'
        endif
    endif

    return '❌ Not loaded'
endfunction

function! llama#setup()
    command! LlamaEnable         call llama#enable()
    command! LlamaDisable        call llama#disable()
    command! LlamaToggle         call llama#toggle()
    command! LlamaToggleAutoFim  call llama#toggle_auto_fim()
    command! LlamaStatus         call llama#status()

    command! -range=% LlamaInstruct call llama#inst(<line1>, <line2>)

    call llama#debug_setup()
endfunction

function! llama#init()
    call llama#debug_log('llama.vim initializing ...')

    if !executable('curl')
        echohl WarningMsg
        echo 'llama.vim requires the "curl" command to be available'
        echohl None
        return
    endif

    call llama#setup()

    let s:fim_data = {}

    let s:ring_chunks = [] " current set of chunks used as extra context
    let s:ring_queued = [] " chunks that are queued to be sent for processing
    let s:ring_n_evict = 0

    let s:fim_hint_shown = v:false
    let s:pos_y_pick = -9999 " last y where we picked a chunk
    let s:indent_last = -1   " last indentation level that was accepted (TODO: this might be buggy)

    let s:timer_fim = -1
    let s:t_last_move = reltime() " last time the cursor moved

    let s:current_job_fim  = v:null

    let s:inst_reqs = {}
    let s:inst_req_id = 0

    let s:ghost_text_nvim = exists('*nvim_buf_get_mark')
    let s:ghost_text_vim = has('textprop')

    if s:ghost_text_vim
        if version < 901
            echom 'Warning: llama.vim requires version 901 or greater. Current version: ' . version
        endif
        let s:hlgroup_hint = 'llama_hl_fim_hint'
        let s:hlgroup_info = 'llama_hl_fim_info'

        let s:hlgroup_inst      = 'llama_hl_inst_src'
        let s:hlgroup_inst_info = 'llama_hl_inst_info'

        if empty(prop_type_get(s:hlgroup_hint))
            call prop_type_add(s:hlgroup_hint, {'highlight': s:hlgroup_hint})
        endif
        if empty(prop_type_get(s:hlgroup_info))
            call prop_type_add(s:hlgroup_info, {'highlight': s:hlgroup_info})
        endif
        if empty(prop_type_get(s:hlgroup_inst))
            call prop_type_add(s:hlgroup_inst, {'highlight': s:hlgroup_inst})
        endif

        if !hlexists(s:hlgroup_inst_info)
            highlight link llama_hl_inst_info Comment
        endif
        if empty(prop_type_get(s:hlgroup_inst_info))
            call prop_type_add(s:hlgroup_inst_info, {'highlight': s:hlgroup_inst_info})
        endif
    endif

    if g:llama_config.enable_at_startup
        call llama#enable()
    endif
endfunction

function! llama#setup_autocmds()
    augroup llama
        autocmd!
        autocmd InsertLeavePre  * call llama#fim_hide()

        autocmd CompleteChanged * call llama#fim_hide()
        autocmd CompleteDone    * call s:on_move()

        if g:llama_config.auto_fim
            autocmd CursorMoved     * call s:on_move()
            autocmd CursorMovedI    * call s:on_move()

            autocmd CursorMovedI * call llama#fim(-1, -1, v:true, [], v:true)
        endif

        " gather chunks upon yanking
        autocmd TextYankPost    * if v:event.operator ==# 'y' | call s:pick_chunk(v:event.regcontents, v:false, v:true) | endif

        " gather chunks upon entering/leaving a buffer
        autocmd BufEnter        * call timer_start(100, {-> s:pick_chunk(getline(max([1, line('.') - g:llama_config.ring_chunk_size/2]), min([line('.') + g:llama_config.ring_chunk_size/2, line('$')])), v:true, v:true)})
        autocmd BufLeave        * call                      s:pick_chunk(getline(max([1, line('.') - g:llama_config.ring_chunk_size/2]), min([line('.') + g:llama_config.ring_chunk_size/2, line('$')])), v:true, v:true)

        " gather chunk upon saving the file
        autocmd BufWritePost    * call s:pick_chunk(getline(max([1, line('.') - g:llama_config.ring_chunk_size/2]), min([line('.') + g:llama_config.ring_chunk_size/2, line('$')])), v:true, v:true)
    augroup END
endfunction

function! llama#enable()
    if s:llama_enabled
        return
    endif

    " setup keymaps
    if g:llama_config.keymap_fim_trigger != ''
        exe "autocmd InsertEnter * inoremap <buffer> <expr> <silent> " . g:llama_config.keymap_fim_trigger . " llama#fim_inline(v:false, v:false)"
    endif
    if g:llama_config.keymap_debug_toggle != ''
        exe "nnoremap <silent> " .. g:llama_config.keymap_debug_toggle .. " :call llama#debug_toggle()<CR>"
    endif

    if g:llama_config.keymap_inst_trigger != ''
        exe "vnoremap <silent> " .. g:llama_config.keymap_inst_trigger  .. " :LlamaInstruct<CR>"
    endif
    if g:llama_config.keymap_inst_rerun != ''
        exe "nnoremap <silent> " .. g:llama_config.keymap_inst_rerun    .. " :call llama#inst_rerun()<CR>"
    endif
    if g:llama_config.keymap_inst_continue != ''
        exe "nnoremap <silent> " .. g:llama_config.keymap_inst_continue .. " :call llama#inst_continue()<CR>"
    endif
    if g:llama_config.keymap_inst_accept != ''
        exe "nnoremap <silent> " .. g:llama_config.keymap_inst_accept   .. " :call llama#inst_accept()<CR>"
    endif
    if g:llama_config.keymap_inst_cancel != ''
        exe "nnoremap <silent> " .. g:llama_config.keymap_inst_cancel   .. " :call llama#inst_cancel()<CR>"
    endif

    call llama#setup_autocmds()

    silent! call llama#fim_hide()

    " init background update of the ring buffer
    if g:llama_config.ring_n_chunks > 0
        call s:ring_update()
    endif

    let s:llama_enabled = v:true

    call llama#debug_log('plugin enabled')
endfunction

" compute how similar two chunks of text are
" 0 - no similarity, 1 - high similarity
function! s:chunk_sim(c0, c1)
    let l:tokens0 = split(join(a:c0, "\n"), '\W\+')
    let l:tokens1 = split(join(a:c1, "\n"), '\W\+')

    let l:set0 = {}
    for l:tok in l:tokens0
        let l:set0[l:tok] = 1
    endfor

    let l:common = 0
    for l:tok in l:tokens1
        if has_key(l:set0, l:tok)
            let l:common += 1
        endif
    endfor

    if (len(l:tokens0) + len(l:tokens1)) == 0
        let l:res = 1.0
    else
        let l:res = 2.0 * l:common / (len(l:tokens0) + len(l:tokens1))
    endif

    "call llama#debug_log('chunk_sim: ' . l:res, join(a:c0, "\n") . '\n\n====================\n\n' . join(a:c1, "\n"))

    return l:res
endfunction

" pick a random chunk of size g:llama_config.ring_chunk_size from the provided text and queue it for processing
"
" no_mod   - do not pick chunks from buffers with pending changes
" do_evict - evict chunks that are very similar to the new one
"
function! s:pick_chunk(text, no_mod, do_evict)
    " do not pick chunks from buffers with pending changes or buffers that are not files
    if a:no_mod && (getbufvar(bufnr('%'), '&modified') || !buflisted(bufnr('%')) || !filereadable(expand('%')))
        return
    endif

    " if the extra context option is disabled - do nothing
    if g:llama_config.ring_n_chunks <= 0
        return
    endif

    " don't pick very small chunks
    if len(a:text) < 3
        return
    endif

    if len(a:text) + 1 < g:llama_config.ring_chunk_size
        let l:chunk = a:text
    else
        let l:l0 = s:rand(0, max([0, len(a:text) - g:llama_config.ring_chunk_size/2]))
        let l:l1 = min([l:l0 + g:llama_config.ring_chunk_size/2, len(a:text)])

        let l:chunk = a:text[l:l0:l:l1]
    endif

    let l:chunk_str = join(l:chunk, "\n") . "\n"

    " check if this chunk is already added
    let l:exist = v:false

    for i in range(len(s:ring_chunks))
        if s:ring_chunks[i].data == l:chunk
            let l:exist = v:true
            break
        endif
    endfor

    for i in range(len(s:ring_queued))
        if s:ring_queued[i].data == l:chunk
            let l:exist = v:true
            break
        endif
    endfor

    if l:exist
        return
    endif

    " evict queued chunks that are very similar to the new one
    for i in range(len(s:ring_queued) - 1, 0, -1)
        if s:chunk_sim(s:ring_queued[i].data, l:chunk) > 0.9
            if a:do_evict
                call remove(s:ring_queued, i)
                let s:ring_n_evict += 1
            else
                return
            endif
        endif
    endfor

    " also from s:ring_chunks
    for i in range(len(s:ring_chunks) - 1, 0, -1)
        if s:chunk_sim(s:ring_chunks[i].data, l:chunk) > 0.9
            if a:do_evict
                call remove(s:ring_chunks, i)
                let s:ring_n_evict += 1
            else
                return
            endif
        endif
    endfor

    " TODO: become parameter ?
    if len(s:ring_queued) == 16
        call remove(s:ring_queued, 0)
    endif

    call add(s:ring_queued, {'data': l:chunk, 'str': l:chunk_str, 'time': reltime(), 'filename': expand('%')})

    "let &statusline = 'extra context: ' . len(s:ring_chunks) . ' / ' . len(s:ring_queued)
endfunction

function! s:ring_get_extra()
    " extra context
    let l:extra = []
    for l:chunk in s:ring_chunks
        call add(l:extra, {
            \ 'text':     l:chunk.str,
            \ 'time':     l:chunk.time,
            \ 'filename': l:chunk.filename
            \ })
    endfor

    return l:extra
endfunction

" picks a queued chunk, sends it for processing and adds it to s:ring_chunks
" called every g:llama_config.ring_update_ms
function! s:ring_update()
    call timer_start(g:llama_config.ring_update_ms, {-> s:ring_update()})

    " skip update if not in normal model and the cursor has moved recently
    if mode() !=# 'n' && reltimefloat(reltime(s:t_last_move)) < 3.0
        return
    endif

    if len(s:ring_queued) == 0
        return
    endif

    " move the first queued chunk to the ring buffer
    if len(s:ring_chunks) == g:llama_config.ring_n_chunks
        call remove(s:ring_chunks, 0)
    endif

    call add(s:ring_chunks, remove(s:ring_queued, 0))

    "let &statusline = 'updated context: ' . len(s:ring_chunks) . ' / ' . len(s:ring_queued)

    " send asynchronous job with the new extra context so that it is ready for the next FIM
    let l:extra = s:ring_get_extra()

    " no samplers needed here
    let l:request = {
        \ 'id_slot':          0,
        \ 'input_prefix':     "",
        \ 'input_suffix':     "",
        \ 'input_extra':      l:extra,
        \ 'prompt':           "",
        \ 'n_predict':        0,
        \ 'temperature':      0.0,
        \ 'samplers':         [],
        \ 'stream':           v:false,
        \ 'cache_prompt':     v:true,
        \ 't_max_prompt_ms':  1,
        \ 't_max_predict_ms': 1,
        \ 'response_fields':  [""]
        \ }

    let l:curl_command = [
        \ "curl",
        \ "--silent",
        \ "--no-buffer",
        \ "--request", "POST",
        \ "--url", g:llama_config.endpoint_fim,
        \ "--header", "Content-Type: application/json",
        \ "--data", "@-",
        \ ]

    if exists ("g:llama_config.model_fim") && len(g:llama_config.model_fim) > 0
        let l:request['model'] = g:llama_config.model_fim
    end

    if exists ("g:llama_config.api_key") && len(g:llama_config.api_key) > 0
        call extend(l:curl_command, ['--header', 'Authorization: Bearer ' .. g:llama_config.api_key])
    endif

    " no callbacks because we don't need to process the response
    let l:request_json = json_encode(l:request)
    if s:ghost_text_nvim
        let jobid = jobstart(l:curl_command, {})
        call chansend(jobid, l:request_json)
        call chanclose(jobid, 'stdin')
    elseif s:ghost_text_vim
        let jobid = job_start(l:curl_command, {})
        let channel = job_getchannel(jobid)
        call ch_sendraw(channel, l:request_json)
        call ch_close_in(channel)
    endif
endfunction

" =====================================
" Fill-in-Middle (FIM) completion
" =====================================

" get the local context at a specified position
" a:prev can optionally contain a previous completion for this position
"   in such cases, create the local context as if the completion was already inserted
function! s:fim_ctx_local(pos_x, pos_y, prev)
    let l:max_y = line('$')

    if empty(a:prev)
        let l:line_cur = getline(a:pos_y)

        let l:line_cur_prefix = strpart(l:line_cur, 0, a:pos_x)
        let l:line_cur_suffix = strpart(l:line_cur, a:pos_x)

        let l:lines_prefix = getline(max([1, a:pos_y - g:llama_config.n_prefix]), a:pos_y - 1)
        let l:lines_suffix = getline(a:pos_y + 1, min([l:max_y, a:pos_y + g:llama_config.n_suffix]))

        " special handling of lines full of whitespaces - start from the beginning of the line
        if match(l:line_cur, '^\s*$') >= 0
            let l:indent = 0

            let l:line_cur_prefix = ""
            let l:line_cur_suffix = ""
        else
            " the indentation of the current line
            let l:indent = strlen(matchstr(l:line_cur, '^\s*'))
        endif
    else
        if len(a:prev) == 1
            let l:line_cur = getline(a:pos_y) . a:prev[0]
        else
            let l:line_cur = a:prev[-1]
        endif

        let l:line_cur_prefix = l:line_cur
        let l:line_cur_suffix = ""

        let l:lines_prefix = getline(max([1, a:pos_y - g:llama_config.n_prefix + len(a:prev) - 1]), a:pos_y - 1)
        if len(a:prev) > 1
            call add(l:lines_prefix, getline(a:pos_y) . a:prev[0])

            for l:line in a:prev[1:-2]
                call add(l:lines_prefix, l:line)
            endfor
        endif

        let l:lines_suffix = getline(a:pos_y + 1, min([l:max_y, a:pos_y + g:llama_config.n_suffix]))

        let l:indent = s:indent_last
    endif

    let l:prefix = ""
        \ . join(l:lines_prefix, "\n")
        \ . "\n"

    let l:middle = ""
        \ . l:line_cur_prefix

    let l:suffix = ""
        \ . l:line_cur_suffix
        \ . "\n"
        \ . join(l:lines_suffix, "\n")
        \ . "\n"

    let l:res = {}

    let l:res['prefix'] = l:prefix
    let l:res['middle'] = l:middle
    let l:res['suffix'] = l:suffix
    let l:res['indent'] = l:indent

    let l:res['line_cur'] = l:line_cur

    let l:res['line_cur_prefix'] = l:line_cur_prefix
    let l:res['line_cur_suffix'] = l:line_cur_suffix

    return l:res
endfunction

" necessary for 'inoremap <expr>'
function! llama#fim_inline(is_auto, use_cache) abort
    " exit if not enabled
    if !s:llama_enabled
        return ''
    endif

    " we already have a suggestion displayed - hide it
    if s:fim_hint_shown && !a:is_auto
        call llama#fim_hide()
        return ''
    endif

    call llama#fim(-1, -1, a:is_auto, [], a:use_cache)

    return ''
endfunction

" the main FIM call
" takes local context around the cursor and sends it together with the extra context to the server for completion
function! llama#fim(pos_x, pos_y, is_auto, prev, use_cache) abort
    let l:pos_x = a:pos_x
    let l:pos_y = a:pos_y

    if l:pos_x < 0
        let l:pos_x = col('.') - 1
    endif

    if l:pos_y < 0
        let l:pos_y = line('.')
    endif

    " avoid sending repeated requests too fast
    if s:current_job_fim != v:null
        if s:timer_fim != -1
            call timer_stop(s:timer_fim)
            let s:timer_fim = -1
        endif

        let s:timer_fim = timer_start(100, {-> llama#fim(a:pos_x, a:pos_y, v:true, a:prev, a:use_cache)})
        return
    endif

    "if s:fim_hint_shown && empty(a:prev)
    "    return
    "endif

    "let s:t_fim_start = reltime()

    let l:ctx_local = s:fim_ctx_local(l:pos_x, l:pos_y, a:prev)

    let l:prefix = l:ctx_local['prefix']
    let l:middle = l:ctx_local['middle']
    let l:suffix = l:ctx_local['suffix']
    let l:indent = l:ctx_local['indent']

    if a:is_auto && len(l:ctx_local['line_cur_suffix']) > g:llama_config.max_line_suffix
        return
    endif

    let l:t_max_predict_ms = g:llama_config.t_max_predict_ms
    if empty(a:prev)
        " the first request is quick - we will launch a speculative request after this one is displayed
        let l:t_max_predict_ms = 250
    endif

    " compute multiple hashes that can be used to generate a completion for which the
    "   first few lines are missing. this happens when we have scrolled down a bit from where the original
    "   generation was done
    "
    let l:hashes = []

    call add(l:hashes, sha256(l:prefix . l:middle . 'Î' . l:suffix))

    let l:prefix_trim = l:prefix
    for i in range(3)
        let l:prefix_trim = substitute(l:prefix_trim, '^[^\n]*\n', '', '')
        if empty(l:prefix_trim)
            break
        endif

        call add(l:hashes, sha256(l:prefix_trim . l:middle . 'Î' . l:suffix))
    endfor

    " if we already have a cached completion for one of the hashes, don't send a request
    if a:use_cache
        for l:hash in l:hashes
            let l:cached = s:cache_get(l:hash)
            if l:cached isnot v:null && len(l:cached) > 0
                return
            endif
        endfor
    endif

    " TODO: this might be incorrect
    let s:indent_last = l:indent

    " TODO: refactor in a function
    let l:text = getline(max([1, line('.') - g:llama_config.ring_chunk_size/2]), min([line('.') + g:llama_config.ring_chunk_size/2, line('$')]))

    let l:l0 = s:rand(0, max([0, len(l:text) - g:llama_config.ring_chunk_size/2]))
    let l:l1 = min([l:l0 + g:llama_config.ring_chunk_size/2, len(l:text)])

    let l:chunk = l:text[l:l0:l:l1]

    " evict chunks that are very similar to the current context
    " this is needed because such chunks usually distort the completion to repeat what was already there
    for i in range(len(s:ring_chunks) - 1, 0, -1)
        if s:chunk_sim(s:ring_chunks[i].data, l:chunk) > 0.5
            call remove(s:ring_chunks, i)
            let s:ring_n_evict += 1
        endif
    endfor

    let l:extra = s:ring_get_extra()

    let l:request = {
        \ 'id_slot':          0,
        \ 'input_prefix':     l:prefix,
        \ 'input_suffix':     l:suffix,
        \ 'input_extra':      l:extra,
        \ 'prompt':           l:middle,
        \ 'n_predict':        g:llama_config.n_predict,
        \ 'stop':             g:llama_config.stop_strings_fim,
        \ 'n_cmpl':           g:llama_config.n_cmpl,
        \ 'n_indent':         l:indent,
        \ 'top_k':            40,
        \ 'top_p':            0.90,
        \ 'samplers':         ["top_k", "top_p", "infill"],
        \ 'stream':           v:false,
        \ 'cache_prompt':     v:true,
        \ 't_max_prompt_ms':  g:llama_config.t_max_prompt_ms,
        \ 't_max_predict_ms': l:t_max_predict_ms,
        \ 'response_fields':  [
        \                       "content",
        \                       "timings/prompt_n",
        \                       "timings/prompt_ms",
        \                       "timings/prompt_per_token_ms",
        \                       "timings/prompt_per_second",
        \                       "timings/predicted_n",
        \                       "timings/predicted_ms",
        \                       "timings/predicted_per_token_ms",
        \                       "timings/predicted_per_second",
        \                       "truncated",
        \                       "tokens_cached",
        \                     ],
        \ }

    let l:curl_command = [
        \ "curl",
        \ "--silent",
        \ "--no-buffer",
        \ "--request", "POST",
        \ "--url", g:llama_config.endpoint_fim,
        \ "--header", "Content-Type: application/json",
        \ "--data", "@-",
        \ ]

    if exists ("g:llama_config.model_fim") && len(g:llama_config.model_fim) > 0
        let l:request['model'] = g:llama_config.model_fim
    end

    if exists ("g:llama_config.api_key") && len(g:llama_config.api_key) > 0
        call extend(l:curl_command, ['--header', 'Authorization: Bearer ' .. g:llama_config.api_key])
    endif

    if s:current_job_fim != v:null
        if s:ghost_text_nvim
            call jobstop(s:current_job_fim)
        elseif s:ghost_text_vim
            call job_stop(s:current_job_fim)
        endif
    endif

    " send the request asynchronously
    let l:request_json = json_encode(l:request)
    if s:ghost_text_nvim
        let s:current_job_fim = jobstart(l:curl_command, {
            \ 'on_stdout': function('s:fim_on_response', [l:hashes]),
            \ 'on_exit':   function('s:fim_on_exit'),
            \ 'stdout_buffered': v:true
            \ })
        call chansend(s:current_job_fim, l:request_json)
        call chanclose(s:current_job_fim, 'stdin')
    elseif s:ghost_text_vim
        let s:current_job_fim = job_start(l:curl_command, {
            \ 'out_cb':    function('s:fim_on_response', [l:hashes]),
            \ 'exit_cb':   function('s:fim_on_exit')
            \ })

        let channel = job_getchannel(s:current_job_fim)
        call ch_sendraw(channel, l:request_json)
        call ch_close_in(channel)
    endif

    " TODO: per-file location
    let l:delta_y = abs(l:pos_y - s:pos_y_pick)

    " gather some extra context nearby and process it in the background
    " only gather chunks if the cursor has moved a lot
    " TODO: something more clever? reranking?
    if a:is_auto && l:delta_y > 32
        let l:max_y = line('$')

        " expand the prefix even further
        call s:pick_chunk(getline(max([1,       l:pos_y - g:llama_config.ring_scope]), max([1,       l:pos_y - g:llama_config.n_prefix])), v:false, v:false)

        " pick a suffix chunk
        call s:pick_chunk(getline(min([l:max_y, l:pos_y + g:llama_config.n_suffix]),   min([l:max_y, l:pos_y + g:llama_config.n_suffix + g:llama_config.ring_chunk_size])), v:false, v:false)

        let s:pos_y_pick = l:pos_y
    endif
endfunction

" callback that processes the FIM result from the server
function! s:fim_on_response(hashes, job_id, data, event = v:null)
    if s:ghost_text_nvim
        let l:raw = join(a:data, "\n")
    elseif s:ghost_text_vim
        let l:raw = a:data
    endif

    " ignore empty results
    if len(l:raw) == 0
        return
    endif

    " ensure the response is valid JSON, starting with a fast check before full decode
    " n_cmpl == 1 returns a single object {"content": ...}, n_cmpl > 1 returns an array [{"content": ...}, ...]
    if (l:raw !~# '^\s*{' && l:raw !~# '^\s*\[') || l:raw !~# '\v"content"\s*:"'
        return
    endif
    try
        let l:decoded = json_decode(l:raw)
    catch
        return
    endtry

    " normalize to list of response objects
    if type(l:decoded) == v:t_dict
        let l:responses = [l:decoded]
    else
        let l:responses = l:decoded
    endif

    " insert each response into the cache ring buffer
    for l:hash in a:hashes
        for l:resp in l:responses
            call s:cache_insert(l:hash, l:resp)
        endfor
    endfor

    " if nothing is currently displayed - show the hint directly
    if !s:fim_hint_shown || !s:fim_data['can_accept']
        " log only non-speculative fims for now
        call llama#debug_log('fim_on_response', get(l:responses[0], 'content', ''))

        let l:pos_x = col('.') - 1
        let l:pos_y = line('.')

        call s:fim_try_hint(l:pos_x, l:pos_y)
    endif
endfunction

function! s:fim_on_exit(job_id, exit_code, event = v:null)
    if a:exit_code != 0
        echom "FIM job failed with exit code: " . a:exit_code
    endif

    let s:current_job_fim = v:null
endfunction

function! s:on_move()
    let s:t_last_move = reltime()

    call llama#fim_hide()

    let l:pos_x = col('.') - 1
    let l:pos_y = line('.')

    call s:fim_try_hint(l:pos_x, l:pos_y)
endfunction

" try to generate a suggestion using the data in the cache
function! s:fim_try_hint(pos_x, pos_y)
    " show the suggestion only in insert mode
    if mode() !~# '\v^(i|ic|ix)$'
        return
    endif

    let l:pos_x = a:pos_x
    let l:pos_y = a:pos_y

    let l:ctx_local = s:fim_ctx_local(l:pos_x, l:pos_y, [])

    let l:prefix = l:ctx_local['prefix']
    let l:middle = l:ctx_local['middle']
    let l:suffix = l:ctx_local['suffix']

    " Phase 1: exact match at current position
    let l:hash = sha256(l:prefix . l:middle . 'Î' . l:suffix)
    let l:responses = s:cache_get(l:hash)

    if l:responses isnot v:null && len(l:responses) > 0
        " exact match found - render with cycling support
        call s:fim_render(l:pos_x, l:pos_y, l:responses, 0)

        " run async speculative FIM in the background for this position
        if s:fim_hint_shown
            call llama#fim(l:pos_x, l:pos_y, v:true, s:fim_data['content'], v:true)
        endif
        return
    endif

    " Phase 2: nearby match - search for a cached completion whose start matches what was typed
    " only pick the single best match, no cycling
    let l:pm = l:prefix . l:middle
    let l:best_len = 0
    let l:best_resp = v:null
    let l:best_offset = -1

    for i in range(128)
        let l:typed = l:pm[-(1 + i):]
        let l:ctx_new = l:pm[:-(2 + i)] . 'Î' . l:suffix
        let l:hash_new = sha256(l:ctx_new)

        let l:cached = s:cache_get(l:hash_new)
        if l:cached is v:null
            continue
        endif

        " try each cached completion for this nearby key
        for l:resp in l:cached
            let l:content = get(l:resp, 'content', '')
            if len(l:content) <= i
                continue
            endif
            if strpart(l:content, 0, i + 1) !=# l:typed
                continue
            endif

            let l:remainder = strpart(l:content, i + 1)
            if len(l:remainder) > l:best_len
                let l:best_len = len(l:remainder)
                let l:best_resp = copy(l:resp)
                let l:best_resp['content'] = l:remainder
                let l:best_offset = i
            endif
        endfor
    endfor

    if l:best_resp isnot v:null
        " single nearby match found - render without cycling
        call s:fim_render(l:pos_x, l:pos_y, [l:best_resp], 0)

        " run async speculative FIM in the background for this position
        if s:fim_hint_shown
            call llama#fim(l:pos_x, l:pos_y, v:true, s:fim_data['content'], v:true)
        endif
    endif
endfunction

" render a suggestion at the current cursor location
" a:responses  - list of response objects from cache
" a:selected   - index of the currently selected completion
function! s:fim_render(pos_x, pos_y, responses, selected)
    " do not show if there is a completion in progress
    if pumvisible()
        return
    endif

    " clear previous virtual text (needed when cycling completions)
    let l:bufnr = bufnr('%')
    if s:ghost_text_nvim
        let l:id_vt_fim = nvim_create_namespace('vt_fim')
        call nvim_buf_clear_namespace(l:bufnr, l:id_vt_fim, 0, -1)
    elseif s:ghost_text_vim
        call prop_remove({'type': s:hlgroup_hint, 'all': v:true})
        call prop_remove({'type': s:hlgroup_info, 'all': v:true})
    endif

    let l:response = a:responses[a:selected]

    let l:can_accept = v:true
    let l:has_info   = v:false

    let l:n_prompt    = 0
    let l:t_prompt_ms = 1.0
    let l:s_prompt    = 0

    let l:n_predict    = 0
    let l:t_predict_ms = 1.0
    let l:s_predict    = 0

    let l:content = []

    " get the generated suggestion
    for l:part in split(get(l:response, 'content', ''), "\n", 1)
        call add(l:content, l:part)
    endfor

    " remove trailing new lines
    while len(l:content) > 0 && l:content[-1] == ""
        call remove(l:content, -1)
    endwhile

    let l:n_cached  = get(l:response, 'tokens_cached', 0)
    let l:truncated = get(l:response, 'timings/truncated', v:false)

    " if response.timings is available
    if has_key(l:response, 'timings/prompt_n') && has_key(l:response, 'timings/prompt_ms') && has_key(l:response, 'timings/prompt_per_second')
        \ && has_key(l:response, 'timings/predicted_n') && has_key(l:response, 'timings/predicted_ms') && has_key(l:response, 'timings/predicted_per_second')
        let l:n_prompt    = get(l:response, 'timings/prompt_n', 0)
        let l:t_prompt_ms = str2float(get(l:response, 'timings/prompt_ms', '1.0'))
        let l:s_prompt    = str2float(get(l:response, 'timings/prompt_per_second', '0.0'))

        let l:n_predict    = get(l:response, 'timings/predicted_n', 0)
        let l:t_predict_ms = str2float(get(l:response, 'timings/predicted_ms', '1.0'))
        let l:s_predict    = str2float(get(l:response, 'timings/predicted_per_second', '0.0'))
    endif

    let l:has_info = v:true

    if len(l:content) == 0
        call add(l:content, "")
        let l:can_accept = v:false
    endif

    let l:pos_x = a:pos_x
    let l:pos_y = a:pos_y

    let l:line_cur = getline(l:pos_y)

    " if the current line is full of whitespaces, trim as much whitespaces from the suggestion
    if match(l:line_cur, '^\s*$') >= 0
        let l:lead = min([strlen(matchstr(l:content[0], '^\s*')), strlen(l:line_cur)])

        let l:line_cur   = strpart(l:content[0], 0, l:lead)
        let l:content[0] = strpart(l:content[0],    l:lead)
    endif

    let l:line_cur_prefix = strpart(l:line_cur, 0, l:pos_x)
    let l:line_cur_suffix = strpart(l:line_cur, l:pos_x)

    " NOTE: the following is logic for discarding predictions that repeat existing text
    "       the code is quite ugly and there is very likely a simpler and more canonical way to implement this
    "
    "       still, I wonder if there is some better way that avoids having to do these special hacks?
    "       on one hand, the LLM 'sees' the contents of the file before we start editing, so it is normal that it would
    "       start generating whatever we have given it via the extra context. but on the other hand, it's not very
    "       helpful to re-generate the same code that is already there

    " truncate the suggestion if the first line is empty
    if len(l:content) == 1 && l:content[0] == ""
        let l:content = [""]
    endif

    " ... and the next lines are repeated
    if len(l:content) > 1 && l:content[0] == "" && l:content[1:] == getline(l:pos_y + 1, l:pos_y + len(l:content) - 1)
        let l:content = [""]
    endif

    " truncate the suggestion if it repeats the suffix
    if len(l:content) == 1 && l:content[0] == l:line_cur_suffix
        let l:content = [""]
    endif

    " find the first non-empty line (strip whitespace)
    let l:cmp_y = l:pos_y + 1
    while l:cmp_y < line('$') && getline(l:cmp_y) =~? '^\s*$'
        let l:cmp_y += 1
    endwhile

    if (l:line_cur_prefix . l:content[0]) == getline(l:cmp_y)
        " truncate the suggestion if it repeats the next line
        if len(l:content) == 1
            let l:content = [""]
        endif

        " ... or if the second line of the suggestion is the prefix of line l:cmp_y + 1
        if len(l:content) == 2 && l:content[-1] == getline(l:cmp_y + 1)[:len(l:content[-1]) - 1]
            let l:content = [""]
        endif

        " ... or if the middle chunk of lines of the suggestion is the same as [l:cmp_y + 1, l:cmp_y + len(l:content) - 1)
        if len(l:content) > 2 && join(l:content[1:-1], "\n") == join(getline(l:cmp_y + 1, l:cmp_y + len(l:content) - 1), "\n")
            let l:content = [""]
        endif
    endif

    " keep only lines that have the same or larger whitespace prefix as l:line_cur_prefix
    "let l:indent = strlen(matchstr(l:line_cur_prefix, '^\s*'))
    "for i in range(1, len(l:content) - 1)
    "    if strlen(matchstr(l:content[i], '^\s*')) < l:indent
    "        let l:content = l:content[:i - 1]
    "        break
    "    endif
    "endfor

    let l:content[-1] .= l:line_cur_suffix

    " if only whitespaces - do not accept
    if join(l:content, "\n") =~? '^\s*$'
        let l:can_accept = v:false
    endif

    let l:info = ''

    " construct the info message
    if g:llama_config.show_info > 0 && l:has_info
        let l:prefix = '   '

        " append completion index when multiple completions are available
        let l:cmpl_idx = ''
        if len(a:responses) > 1
            let l:cmpl_idx = printf(' [%d/%d]', a:selected + 1, len(a:responses))
        endif

        if l:truncated
            let l:info = printf("%s | WARNING: the context is full: %d, increase the server context size or reduce g:llama_config.ring_n_chunks%s",
                \ g:llama_config.show_info == 2 ? l:prefix : 'llama.vim',
                \ l:n_cached,
                \ l:cmpl_idx
                \ )
        else
            let l:info = printf("%s | c: %d, r: %d/%d, e: %d, q: %d/16, C: %d | p: %d (%.2f ms, %.2f t/s) | g: %d (%.2f ms, %.2f t/s)%s",
                \ g:llama_config.show_info == 2 ? l:prefix : 'llama.vim',
                \ l:n_cached,  len(s:ring_chunks), g:llama_config.ring_n_chunks, s:ring_n_evict, len(s:ring_queued),
                \ s:cache_count(),
                \ l:n_prompt,  l:t_prompt_ms,  l:s_prompt,
                \ l:n_predict, l:t_predict_ms, l:s_predict,
                \ l:cmpl_idx
                \ )
        endif

        if g:llama_config.show_info == 1
            " display the info in the statusline
            let &statusline = l:info
            let l:info = ''
        endif
    endif

    " display the suggestion and append the info to the end of the first line
    if s:ghost_text_nvim
        call nvim_buf_set_extmark(l:bufnr, l:id_vt_fim, l:pos_y - 1, l:pos_x, {
            \ 'virt_text': [[l:content[0], 'llama_hl_fim_hint'], [l:info, 'llama_hl_fim_info']],
            \ 'virt_text_pos': l:content == [""] ? 'eol' : 'overlay'
            \ })

        call nvim_buf_set_extmark(l:bufnr, l:id_vt_fim, l:pos_y - 1, 0, {
            \ 'virt_lines': map(l:content[1:], {idx, val -> [[val, 'llama_hl_fim_hint']]})
            \ })
    elseif s:ghost_text_vim
        let l:full_suffix = l:content[0]
        if !empty(l:full_suffix)
            let l:new_suffix = l:full_suffix[0:-len(l:line_cur[l:pos_x:])-1]
            call prop_add(l:pos_y, l:pos_x + 1, {
                \ 'type': s:hlgroup_hint,
                \ 'text': l:new_suffix
                \ })
        endif
        for line in l:content[1:]
            call prop_add(l:pos_y, 0, {
                \ 'type': s:hlgroup_hint,
                \ 'text': line,
                \ 'text_padding_left': s:get_indent(line),
                \ 'text_align': 'below'
                \ })
        endfor
        if !empty(l:info)
            call prop_add(l:pos_y, 0, {
                \ 'type': s:hlgroup_info,
                \ 'text': l:info,
                \ 'text_wrap': 'truncate'
                \ })
        endif
    endif

    " setup accept shortcuts
    if g:llama_config.keymap_fim_accept_full != ''
        exe 'inoremap <buffer> ' . g:llama_config.keymap_fim_accept_full . ' <C-O>:call llama#fim_accept(''full'')<CR>'
    endif
    if g:llama_config.keymap_fim_accept_line != ''
        exe 'inoremap <buffer> ' . g:llama_config.keymap_fim_accept_line . ' <C-O>:call llama#fim_accept(''line'')<CR>'
    endif
    if g:llama_config.keymap_fim_accept_word != ''
        exe 'inoremap <buffer> ' . g:llama_config.keymap_fim_accept_word . ' <C-O>:call llama#fim_accept(''word'')<CR>'
    endif

    " setup cycle shortcuts (always, to prevent <C-J>/<C-K> from moving the cursor)
    " llama#fim_cycle returns '' early when there is nothing to cycle
    if g:llama_config.keymap_fim_next != ''
        exe 'inoremap <expr> <buffer> ' . g:llama_config.keymap_fim_next . ' llama#fim_cycle(1)'
    endif
    if g:llama_config.keymap_fim_prev != ''
        exe 'inoremap <expr> <buffer> ' . g:llama_config.keymap_fim_prev . ' llama#fim_cycle(-1)'
    endif

    let s:fim_hint_shown = v:true

    let s:fim_data['pos_x']       = l:pos_x
    let s:fim_data['pos_y']       = l:pos_y

    let s:fim_data['line_cur']    = l:line_cur

    let s:fim_data['can_accept']  = l:can_accept
    let s:fim_data['content']     = l:content
    let s:fim_data['responses']   = a:responses
    let s:fim_data['selected']    = a:selected
endfunction

" if accept_type == 'full', accept entire response
" if accept_type == 'line', accept only the first line of the response
" if accept_type == 'word', accept only the first word of the response
function! llama#fim_accept(accept_type)
    let l:pos_x  = s:fim_data['pos_x']
    let l:pos_y  = s:fim_data['pos_y']

    let l:line_cur = s:fim_data['line_cur']

    let l:can_accept = s:fim_data['can_accept']
    let l:content    = s:fim_data['content']

    if l:can_accept && len(l:content) > 0
        " insert suggestion on current line
        if a:accept_type != 'word'
            " insert first line of suggestion
            call setline(l:pos_y, l:line_cur[:(l:pos_x - 1)] . l:content[0])
        else
            " insert first word of suggestion
            let l:suffix = l:line_cur[(l:pos_x):]
            let l:word = matchstr(l:content[0][:-(len(l:suffix) + 1)], '^\s*\S\+')
            call setline(l:pos_y, l:line_cur[:(l:pos_x - 1)] . l:word . l:suffix)
        endif

        " insert rest of suggestion
        if len(l:content) > 1 && a:accept_type == 'full'
            call append(l:pos_y, l:content[1:-1])
        endif

        " move cusor
        if a:accept_type == 'word'
            " move cursor to end of word
            call cursor(l:pos_y, l:pos_x + len(l:word) + 1)
        elseif a:accept_type == 'line' || len(l:content) == 1
            " move cursor for 1-line suggestion
            call cursor(l:pos_y, l:pos_x + len(l:content[0]) + 1)
            if len(l:content) > 1
                " simulate pressing Enter to move to next line
                call feedkeys("\<CR>")
            endif
        else
            " move cursor for multi-line suggestion
            call cursor(l:pos_y + len(l:content) - 1, len(l:content[-1]) + 1)
        endif
    endif

    call llama#fim_hide()
endfunction

function! llama#fim_hide()
    let s:fim_hint_shown = v:false

    " clear the virtual text
    let l:bufnr = bufnr('%')

    if s:ghost_text_nvim
        let l:id_vt_fim = nvim_create_namespace('vt_fim')
        call nvim_buf_clear_namespace(l:bufnr, l:id_vt_fim,  0, -1)
    elseif s:ghost_text_vim
        call prop_remove({'type': s:hlgroup_hint, 'all': v:true})
        call prop_remove({'type': s:hlgroup_info, 'all': v:true})
    endif

    " Clear the statusline if show_info was set to 1
    if g:llama_config.show_info == 1
        set statusline=
    endif

    " remove the mappings
    if g:llama_config.keymap_fim_accept_full != ''
        exe 'silent! iunmap <buffer> ' . g:llama_config.keymap_fim_accept_full
    endif
    if g:llama_config.keymap_fim_accept_line != ''
        exe 'silent! iunmap <buffer> ' . g:llama_config.keymap_fim_accept_line
    endif
    if g:llama_config.keymap_fim_accept_word != ''
        exe 'silent! iunmap <buffer> ' . g:llama_config.keymap_fim_accept_word
    endif
    if g:llama_config.keymap_fim_next != ''
        exe 'silent! iunmap <buffer> ' . g:llama_config.keymap_fim_next
    endif
    if g:llama_config.keymap_fim_prev != ''
        exe 'silent! iunmap <buffer> ' . g:llama_config.keymap_fim_prev
    endif
endfunction

" ref: https://github.com/ggml-org/llama.vim/pull/85
function! llama#is_fim_hint_shown()
    return s:fim_hint_shown
endfunction

" cycle to the next/previous completion
" a:direction - 1 for next, -1 for previous
" returns '' for <expr> mapping to consume the key
function! llama#fim_cycle(direction)
    if !s:fim_hint_shown
        return ''
    endif

    let l:n = len(s:fim_data['responses'])
    if l:n <= 1
        return ''
    endif

    let s:fim_data['selected'] = (s:fim_data['selected'] + a:direction + l:n) % l:n

    call s:fim_render(s:fim_data['pos_x'], s:fim_data['pos_y'],
        \ s:fim_data['responses'], s:fim_data['selected'])

    return ''
endfunction

" =====================================
" Instruct-based editing
" =====================================

function! llama#inst_build(l0, l1, inst, inst_prev = [])
    let l:prefix    = getline(max([1, a:l0 - g:llama_config.n_prefix]), a:l0 - 1)
    let l:selection = getline(a:l0, a:l1)
    let l:suffix    = getline(a:l1 + 1, min ([line('$'), a:l1 + g:llama_config.n_suffix]))

    if !empty(a:inst_prev)
        let l:messages = copy(a:inst_prev)
    else
        let l:system_prompt  = ""
        let l:system_prompt .= "You are a text-editing assistant. Respond ONLY with the result of applying INSTRUCTION to SELECTION given the CONTEXT. Maintain the existing text indentation. Do not add extra code blocks. Respond only with the modified block. If the INSTRUCTION is a question, answer it directly. Do not output any extra separators. Consider the local context before (PREFIX) and after (SUFFIX) the SELECTION.\n"

        let l:extra = s:ring_get_extra()

        " note: this has side effects as it escapes newlines and quotes, which prevents does not work well with context-based speculative approaches
        "let l:payload = {'CONTEXT': join(l:extra, "\n"), 'PREFIX': join(l:prefix, "\n"), 'SELECTION': join(l:selection, "\n"), 'SUFFIX': join(l:suffix, "\n")}
        "let l:system_prompt .= "\n" . json_encode(l:payload) . "\n"

        let l:system_prompt .= "\n"
        let l:system_prompt .= "--- CONTEXT     " . repeat('-', 40) . "\n"
        let l:system_prompt .= join(l:extra, "\n") . "\n"
        let l:system_prompt .= "--- PREFIX      " . repeat('-', 40) . "\n"
        let l:system_prompt .= join(l:prefix, "\n") . "\n"
        let l:system_prompt .= "--- SELECTION   " . repeat('-', 40) . "\n"
        let l:system_prompt .= join(l:selection, "\n") . "\n"
        let l:system_prompt .= "--- SUFFIX      " . repeat('-', 40) . "\n"
        let l:system_prompt .= join(l:suffix, "\n") . "\n"

        let l:system_message = {
            \ 'role': 'system',
            \ 'content': l:system_prompt,
            \ }

        let l:messages = [l:system_message]
    endif

    let l:user_content  = ""

    if !empty(a:inst)
        let l:user_content .= "INSTRUCTION: " . a:inst
    endif

    let l:user_message = {'role': 'user', 'content': l:user_content}

    call add(l:messages, l:user_message)

    return l:messages
endfunction

function! llama#inst(l0, l1)
    let l:l0 = a:l0
    let l:l1 = a:l1

    " create request state
    let l:req_id = s:inst_req_id
    let s:inst_req_id += 1

    " while the user is providing an instruction, send a warm-up request
    let l:messages = llama#inst_build(l:l0, l:l1, '')

    let l:request = {
        \ 'id_slot':      l:req_id,
        \ 'messages':     l:messages,
        \ 'samplers':     [],
        \ 'n_predict':    0,
        \ 'stream':       v:false,
        \ 'stop':         g:llama_config.stop_strings_inst,
        \ 'cache_prompt': v:true,
        \ 'response_fields':  [""],
        \ }

    let l:curl_command = [
        \ "curl",
        \ "--silent",
        \ "--no-buffer",
        \ "--request", "POST",
        \ "--url", g:llama_config.endpoint_inst,
        \ "--header", "Content-Type: application/json",
        \ "--data", "@-",
        \ ]

    if exists("g:llama_config.model_inst") && len("g:llama_config.model_inst") > 0
        let l:request.model = g:llama_config.model_inst
    endif

    if exists("g:llama_config.api_key") && len("g:llama_config.api_key") > 0
        call extend(l:curl_command, ['--header', 'Authorization: Bearer ' .. g:llama_config.api_key])
    endif

    let l:request_json = json_encode(l:request)

    " no callbacks because we don't need to process the response
    if s:ghost_text_nvim
        let jobid = jobstart(l:curl_command, {})
        call chansend(jobid, l:request_json)
        call chanclose(jobid, 'stdin')
    elseif s:ghost_text_vim
        let jobid = job_start(l:curl_command, {})
        let channel = job_getchannel(jobid)
        call ch_sendraw(channel, l:request_json)
        call ch_close_in(channel)
    endif

    let l:inst = input('Instruction: ')
    if empty(l:inst)
        return
    endif

    call llama#debug_log('inst_send | ' . l:inst)

    let l:bufnr = bufnr('%')

    let l:req = {
        \ 'id': l:req_id,
        \ 'bufnr': l:bufnr,
        \ 'range': [l:l0, l:l1],
        \ 'status': 'proc',
        \ 'result': '',
        \ 'inst': l:inst,
        \ 'inst_prev': [],
        \ 'job': v:null,
        \ 'n_gen': 0,
        \ 'extmark': -1,
        \ 'extmark_virt': -1,
        \ }

    let s:inst_reqs[l:req_id] = l:req

    " highlights the selected text
    if s:ghost_text_nvim
        let l:ns = nvim_create_namespace('vt_inst')
        let l:req.extmark = nvim_buf_set_extmark(l:bufnr, l:ns, l:l0 - 1, 0, {
            \ 'end_row': l:l1 - 1,
            \ 'end_col': len(getline(l:l1)),
            \ 'hl_group': 'llama_hl_inst_src'
            \ })
    elseif s:ghost_text_vim
        let l:prop_id = prop_add(l:l0, 1, {
            \ 'type': 'llama_hl_inst_src',
            \ 'end_lnum': l:l1,
            \ 'end_col': len(getline(l:l1)) + 1
            \ })
        let l:req.extmark = l:prop_id
    endif

    " Initialize virtual text with processing status
    call s:inst_update(l:req_id, 'proc')

    let l:req.inst_prev = llama#inst_build(l:l0, l:l1, l:inst)

    call llama#inst_send(l:req_id, l:req.inst_prev)
endfunction

function! llama#inst_send(req_id, messages)
    call llama#debug_log('inst_send', join(a:messages, "\n"))

    let l:request = {
        \ 'id_slot':      a:req_id,
        \ 'messages':     a:messages,
        \ 'min_p':        0.1,
        \ 'temperature':  0.1,
        \ 'samplers':     ["min_p", "temperature"],
        \ 'stop':         g:llama_config.stop_strings_inst,
        \ 'stream':       v:true,
        \ 'cache_prompt': v:true,
        \ }

    let l:curl_command = [
        \ "curl",
        \ "--silent",
        \ "--no-buffer",
        \ "--request", "POST",
        \ "--url", g:llama_config.endpoint_inst,
        \ "--header", "Content-Type: application/json",
        \ "--data", "@-",
        \ ]

    if exists("g:llama_config.model_inst") && len("g:llama_config.model_inst") > 0
        let l:request.model = g:llama_config.model_inst
    endif

    if exists("g:llama_config.api_key") && len("g:llama_config.api_key") > 0
        call extend(l:curl_command, ['--header', 'Authorization: Bearer ' .. g:llama_config.api_key])
    endif

    let l:request_json = json_encode(l:request)

    let l:req = s:inst_reqs[a:req_id]

    if s:ghost_text_nvim
        let l:req.job = jobstart(l:curl_command, {
            \ 'on_stdout': function('s:inst_on_response', [a:req_id]),
            \ 'on_exit':   function('s:inst_on_exit',     [a:req_id]),
            \ 'stdout_buffered': v:false
            \ })
        call chansend(l:req.job, l:request_json)
        call chanclose(l:req.job, 'stdin')
    elseif s:ghost_text_vim
        let l:req.job = job_start(l:curl_command, {
            \ 'out_cb':  function('s:inst_on_response', [a:req_id]),
            \ 'exit_cb': function('s:inst_on_exit',     [a:req_id])
            \ })

        let channel = job_getchannel(l:req.job)
        call ch_sendraw(channel, l:request_json)
        call ch_close_in(channel)
    endif
endfunction

function! llama#inst_update_pos(req)
    let l:bufnr = a:req.bufnr

    if s:ghost_text_nvim
        let l:ns = nvim_create_namespace('vt_inst')

        let l:extmark_pos = nvim_buf_get_extmark_by_id(l:bufnr, l:ns, a:req.extmark, {})
        if empty(l:extmark_pos)
            continue
        endif

        let l:extmark_line = l:extmark_pos[0] + 1
        let a:req.range[1] = l:extmark_line + a:req.range[1] - a:req.range[0]
        let a:req.range[0] = l:extmark_line
    else
        " TODO: implement classic Vim support
    endif
endfunction

function! s:inst_update(id, status)
    if !has_key(s:inst_reqs, a:id)
        return
    endif

    let l:req = s:inst_reqs[a:id]

    let l:req.status = a:status
    call llama#inst_update_pos(l:req)

    if s:ghost_text_nvim
        let l:ns = nvim_create_namespace('vt_inst')

        if l:req.extmark_virt != -1
            call nvim_buf_del_extmark(l:req.bufnr, l:ns, l:req.extmark_virt)
            let l:req.extmark_virt = -1
        endif

        let l:inst_trunc = l:req.inst
        if len(l:inst_trunc) > 128
            let l:inst_trunc = l:inst_trunc[:127] . '...'
        endif

        let l:hl = ''
        let l:sep = '====================================='

        let l:virt_lines = []
        if a:status == 'ready'
            let l:result_lines = split(l:req.result, "\n")

            let l:hl = 'llama_hl_inst_virt_ready'
            let l:virt_lines = [[[l:sep, l:hl]]] + map(l:result_lines, {idx, val -> [[val, l:hl]]})
        elseif a:status == 'proc'
            let l:hl = 'llama_hl_inst_virt_proc'
            let l:virt_lines = [
                \ [[l:sep, l:hl]],
                \ [[printf('Endpoint:    %s', g:llama_config.endpoint_inst), l:hl]],
                \ [[printf('Model:       %s', g:llama_config.model_inst),    l:hl]],
                \ [[printf('Instruction: %s', l:inst_trunc),                 l:hl]],
                \ [[printf('Processing ...'),                                l:hl]]
                \ ]
        elseif a:status == 'gen'
            let l:preview = substitute(l:req.result, '.*\n\s*', '', '')
            if len(l:req.result) == 0
                let l:preview = '[thinking]'
            endif

            let l:hl = 'llama_hl_inst_virt_gen'
            let l:virt_lines = [
                \ [[l:sep, l:hl]],
                \ [[printf('Endpoint:    %s', g:llama_config.endpoint_inst),        l:hl]],
                \ [[printf('Model:       %s', g:llama_config.model_inst),           l:hl]],
                \ [[printf('Instruction: %s', l:inst_trunc),                        l:hl]],
                \ [[printf('Generating:  %4d tokens | %s', l:req.n_gen, l:preview), l:hl]],
                \ ]
        endif

        if !empty(l:virt_lines)
            let l:virt_lines = l:virt_lines + [[[l:sep, l:hl]]]
            let l:req.extmark_virt = nvim_buf_set_extmark(l:req.bufnr, l:ns, l:req.range[1] - 1, 0, {
                \ 'virt_lines': l:virt_lines
                \ })
        endif
    elseif s:ghost_text_vim
        if l:req.extmark_virt != -1
             call prop_remove({
                \ 'type': 'llama_hl_inst_info',
                \ 'id': l:req.extmark_virt,
                \ 'bufnr': l:req.bufnr
                \ })
        endif

        let l:text = ''

        if a:status == 'proc'
            let l:text = ' ⏳ Processing... (' . g:llama_config.model_inst . ')'
        elseif a:status == 'gen'
            let l:preview = substitute(l:req.result, '.*\n\s*', '', '')
            " Truncate if too long so it fits on screen
            if len(l:preview) > 40
                let l:preview = l:preview[:37] . '...'
            endif
            let l:text = printf(' ⏳ Generating (%d tokens): %s', l:req.n_gen, l:preview)
        elseif a:status == 'ready'
            let l:text = ' ✅ Ready! Press <Tab> to accept.'
        endif

        let l:prop_id = 9000 + a:id

        call prop_add(l:req.range[1], 0, {
            \ 'type': 'llama_hl_inst_info',
            \ 'id': l:prop_id,
            \ 'bufnr': l:req.bufnr,
            \ 'text': l:text,
            \ 'text_align': 'after',
            \ 'text_padding_left': 2
            \ })

        let l:req.extmark_virt = l:prop_id
    endif
endfunction

function! s:inst_on_response(id, job_id, data, event = v:null)
    if has('nvim')
        let l:lines = a:data
    else
        let l:lines = [a:data]
    endif

    if len(l:lines) == 0
        return
    endif

    let l:content = ''
    for l:line in l:lines
        if len(l:line) > 5 && l:line[:5] ==# 'data: '
            let l:line = l:line[6:]
        endif

        if empty(l:line) || l:line =~# '^\s*$'
            continue
        endif

        try
            let l:response = json_decode(l:line)
            let l:choices = get(l:response, 'choices', [{}])

            if has_key(l:choices[0], 'delta')
                " stream = true
                let l:delta = l:choices[0].delta
                if has_key(l:delta, 'content')
                    let l:delta = l:delta.content
                    if type(l:delta) == v:t_string
                        let l:content .= l:delta
                    endif
                endif
            elseif has_key(l:choices[0], 'message')
                " stream = false
                let l:delta = l:choices[0].message.content
                if type(l:delta) == v:t_string
                    let l:content .= l:delta
                endif
            endif
        catch
            " non-json
            call llama#debug_log('inst_on_response parse error', l:line)
        endtry
    endfor

    if !has_key(s:inst_reqs, a:id)
        return
    endif

    call s:inst_update(a:id, 'gen')

    let l:req = s:inst_reqs[a:id]

    if !empty(l:content)
        let l:req.result .= l:content
    endif

    let l:req.n_gen = l:req.n_gen + 1
endfunction

function! s:inst_on_exit(id, job_id, exit_code, event = v:null)
    if a:exit_code != 0
        echom "Instruct job failed with exit code: " . a:exit_code
        call s:inst_remove(a:id)
        return
    endif

    if !has_key(s:inst_reqs, a:id)
        return
    endif

    call s:inst_update(a:id, 'ready')

    " add assistant response to messages for continuation
    let l:req = s:inst_reqs[a:id]
    call add(l:req.inst_prev, {'role': 'assistant', 'content': l:req.result})
endfunction

function! s:inst_remove(id)
    if has_key(s:inst_reqs, a:id)
        let l:req = s:inst_reqs[a:id]
        if s:ghost_text_nvim
            let l:ns = nvim_create_namespace('vt_inst')
            call nvim_buf_del_extmark(l:req.bufnr, l:ns, l:req.extmark)
            if l:req.extmark_virt != -1
                call nvim_buf_del_extmark(l:req.bufnr, l:ns, l:req.extmark_virt)
            endif
            call jobstop(l:req.job)
        elseif s:ghost_text_vim
            call prop_remove({
                \ 'type': 'llama_hl_inst_src',
                \ 'bufnr': l:req.bufnr
                \ })

            call prop_remove({
                \ 'type': 'llama_hl_inst_info',
                \ 'bufnr': l:req.bufnr
                \ })

            call job_stop(l:req.job)
        endif

        call remove(s:inst_reqs, a:id)
    endif
endfunction

function! s:inst_callback(bufnr, l0, l1, result)
    let l:result_lines = split(a:result, "\n", 1)

    " Remove trailing empty lines
    while len(l:result_lines) > 0 && l:result_lines[-1] == ""
        call remove(l:result_lines, -1)
    endwhile

    let l:num_result = len(l:result_lines)
    let l:num_original = a:l1 - a:l0 + 1

    " Delete the original range
    call deletebufline(a:bufnr, a:l0, a:l1)

    " Insert the new lines
    call append(a:l0 - 1, l:result_lines)
endfunction

function! llama#inst_accept()
    let l:line = line('.')

    for l:req in values(s:inst_reqs)
        if l:req.status ==# 'ready'
            call llama#inst_update_pos(l:req)

            if l:line >= l:req.range[0] && l:line <= l:req.range[1]
                call s:inst_remove(l:req.id)
                call s:inst_callback(l:req.bufnr, l:req.range[0], l:req.range[1], l:req.result)
                return
            endif
        endif
    endfor

    call feedkeys("\<Tab>", 'n')
endfunction

function! llama#inst_cancel()
    let l:line = line('.')
    for l:req in values(s:inst_reqs)
        if l:line >= l:req.range[0] && l:line <= l:req.range[1]
            call s:inst_remove(l:req.id)
            return
        endif
    endfor
    " If not in range, do normal Esc (nothing)
endfunction

function! llama#inst_rerun()
    let l:lnum = line('.')
    for l:req in values(s:inst_reqs)
        call llama#inst_update_pos(l:req)

        if l:req.status == 'ready' && l:lnum >= l:req.range[0] && l:lnum <= l:req.range[1]
            call llama#debug_log('inst_rerun')

            let l:req.result = ''
            let l:req.status = 'proc'
            let l:req.n_gen = 0

            call remove(l:req.inst_prev, -1)

            call s:inst_update(l:req.id, 'proc')

            call llama#inst_send(l:req.id, l:req.inst_prev)
            return
        endif
    endfor
endfunction

function! llama#inst_continue()
    let l:lnum = line('.')
    for l:req in values(s:inst_reqs)
        call llama#inst_update_pos(l:req)

        if l:req.status == 'ready' && l:lnum >= l:req.range[0] && l:lnum <= l:req.range[1]
            let l:inst = input('Next instruction: ')
            if empty(l:inst)
                return
            endif

            call llama#debug_log('inst_continue | ' . l:inst)

            let l:req.result = ''
            let l:req.status = 'proc'
            let l:req.inst = l:inst
            let l:req.n_gen = 0

            call s:inst_update(l:req.id, 'proc')

            let l:req.inst_prev = llama#inst_build(l:req.range[0], l:req.range[1], l:inst, l:req.inst_prev)

            call llama#inst_send(l:req.id, l:req.inst_prev)
            return
        endif
    endfor
    " If not in active edit, do nothing
endfunction

" =====================================
" Debug helpers
" =====================================

function! llama#debug_log(msg, ...) abort
    return call('llama_debug#log', [a:msg] + a:000)
endfunction

function! llama#debug_toggle() abort
    return llama_debug#toggle()
endfunction

function! llama#debug_clear() abort
    return llama_debug#clear()
endfunction

function! llama#debug_setup() abort
    return llama_debug#setup()
endfunction