llama.vim: filter server response fields (#24)

* filter response fields

* clean up

* response fields for ring update

* formatting change

* Update autoload/llama.vim

formatting comma

Co-authored-by: Georgi Gerganov <[email protected]>

* removed n_ctx

---------

Co-authored-by: Georgi Gerganov <[email protected]>

Author: VJHack

autoload/llama.vim

@@ -307,9 +307,9 @@ function! s:ring_update()
         \ 'samplers':         ["temperature"],
         \ 'cache_prompt':     v:true,
         \ 't_max_prompt_ms':  1,
-        \ 't_max_predict_ms': 1
+        \ 't_max_predict_ms': 1,
+        \ 'response_fields':  [""]
         \ })
-
     let l:curl_command = [
         \ "curl",
         \ "--silent",
@@ -420,7 +420,20 @@ function! llama#fim(is_auto, cache) abort
         \ 'samplers':         ["top_k", "top_p", "infill"],
         \ 'cache_prompt':     v:true,
         \ 't_max_prompt_ms':  g:llama_config.t_max_prompt_ms,
-        \ 't_max_predict_ms': g:llama_config.t_max_predict_ms
+        \ 't_max_predict_ms': g:llama_config.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 = [
@@ -662,24 +675,21 @@ function! s:fim_on_stdout(hash, cache, pos_x, pos_y, is_auto, job_id, data, even
             call remove(s:content, -1)
         endwhile
 
-        let l:generation_settings = get(l:response, 'generation_settings', {})
-        let l:n_ctx = get(l:generation_settings, 'n_ctx', 0)
-
-        let l:n_cached  = get(l:response, 'tokens_cached', 0)
-        let l:truncated = get(l:response, 'truncated', v:false)
+        let l:n_cached  = get(l:response, 'timings/tokens_cached', 0)
+        let l:truncated = get(l:response, 'timings/truncated', v:false)
 
         " if response.timings is available
-        if len(get(l:response, 'timings', {})) > 0
+        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:has_info = v:true
-            let l:timings  = get(l:response, 'timings', {})
 
-            let l:n_prompt    = get(l:timings, 'prompt_n', 0)
-            let l:t_prompt_ms = get(l:timings, 'prompt_ms', 1)
-            let l:s_prompt    = get(l:timings, 'prompt_per_second', 0)
+            let l:n_prompt    = get(l:response, 'timings/prompt_n', 0)
+            let l:t_prompt_ms = get(l:response, 'timings/prompt_ms', 1)
+            let l:s_prompt    = get(l:response, 'timings/prompt_per_second', 0)
 
-            let l:n_predict    = get(l:timings, 'predicted_n', 0)
-            let l:t_predict_ms = get(l:timings, 'predicted_ms', 1)
-            let l:s_predict    = get(l:timings, 'predicted_per_second', 0)
+            let l:n_predict    = get(l:response, 'timings/predicted_n', 0)
+            let l:t_predict_ms = get(l:response, 'timings/predicted_ms', 1)
+            let l:s_predict    = get(l:response, 'timings/predicted_per_second', 0)
         endif
 
         " if response was pulled from cache
@@ -772,9 +782,9 @@ function! s:fim_on_stdout(hash, cache, pos_x, pos_y, is_auto, job_id, data, even
         let l:prefix = '   '
 
         if l:truncated
-            let l:info = printf("%s | WARNING: the context is full: %d / %d, increase the server context size or reduce g:llama_config.ring_n_chunks",
+            let l:info = printf("%s | WARNING: the context is full: %d, increase the server context size or reduce g:llama_config.ring_n_chunks",
                 \ g:llama_config.show_info == 2 ? l:prefix : 'llama.vim',
-                \ l:n_cached, l:n_ctx
+                \ l:n_cached
                 \ )
         elseif l:is_cached
             let l:info = printf("%s | C: %d / %d, | t: %.2f ms",
@@ -783,9 +793,9 @@ function! s:fim_on_stdout(hash, cache, pos_x, pos_y, is_auto, job_id, data, even
                 \ 1000.0 * reltimefloat(reltime(s:t_fim_start))
                 \ )
         else
-            let l:info = printf("%s | c: %d / %d, r: %d / %d, e: %d, q: %d / 16 | p: %d (%.2f ms, %.2f t/s) | g: %d (%.2f ms, %.2f t/s) | t: %.2f ms",
+            let l:info = printf("%s | c: %d, r: %d / %d, e: %d, q: %d / 16 | p: %d (%.2f ms, %.2f t/s) | g: %d (%.2f ms, %.2f t/s) | t: %.2f ms",
                 \ g:llama_config.show_info == 2 ? l:prefix : 'llama.vim',
-                \ l:n_cached,  l:n_ctx, len(s:ring_chunks), g:llama_config.ring_n_chunks, s:ring_n_evict, len(s:ring_queued),
+                \ l:n_cached,  len(s:ring_chunks), g:llama_config.ring_n_chunks, s:ring_n_evict, len(s:ring_queued),
                 \ l:n_prompt,  l:t_prompt_ms,  l:s_prompt,
                 \ l:n_predict, l:t_predict_ms, l:s_predict,
                 \ 1000.0 * reltimefloat(reltime(s:t_fim_start))