`battle/80b-watch.log`

48 lines
[09:37:24] waiting for mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit to SERVE (real probe)...
[09:37:49] dl=13890MB probe=''
[09:38:34] dl=15058MB probe=''
[09:39:19] dl=16194MB probe=''
[09:40:04] dl=17362MB probe=''
[09:40:49] dl=18498MB probe=''
[09:41:34] dl=19682MB probe=''
[09:42:19] dl=20770MB probe=''
[09:43:04] dl=21954MB probe=''
[09:43:49] dl=23122MB probe=''
[09:44:34] dl=24290MB probe=''
[09:45:19] dl=25442MB probe=''
[09:46:04] dl=26610MB probe=''
[09:46:49] dl=27762MB probe=''
[09:47:34] dl=28930MB probe=''
[09:48:19] dl=30050MB probe=''
[09:49:04] dl=31186MB probe=''
[09:49:49] dl=32354MB probe=''
[09:50:34] dl=33506MB probe=''
[09:51:19] dl=34642MB probe=''
[09:52:04] dl=35810MB probe=''
[09:52:49] dl=36914MB probe=''
[09:53:34] dl=38125MB probe=''
[09:54:19] dl=39281MB probe=''
[09:55:04] dl=40449MB probe=''
[09:55:49] dl=41393MB probe=''
[09:56:34] dl=42289MB probe=''
[09:57:20] dl=42815MB probe=''
[09:58:04] dl=42815MB probe='ready'
[09:58:04] SERVING — running eval

=== mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit ===
score 12/12  |  tok/s 21.4  |  longctx placed=true pass=true (~33990 tok)
byCat: json 3/3  classify 3/3  constraint 2/2  reason 2/2  code 2/2
  ✅ json_person    [object Object]
  ✅ json_order     [object Object]
  ✅ json_list      [object Object]
  ✅ cls_sentiment  negative
  ✅ cls_ticket     billing
  ✅ cls_spam       yes
  ✅ cons_oneword   Paris
  ✅ cons_numonly   391
  ✅ reason_word    wall: $ 135 - 108 = 27 $ sq ft. ### ✅ Final Answer: **27**
  ✅ reason_logic   Mia
  ✅ code_sum       function sumTo(n){let s=0;for(let i=1;i<=n;i++)s+=i;return s;}
  ✅ code_null      function total(o){return (o.price || 0) * (o.qty || 0);}
[09:59:04] EVAL DONE — 80b.json written