vh/esh-pfi-infrastructure
1
0
Fork 0
Code Issues 7 Pull Requests Actions Packages Projects Releases Wiki Activity
Files
03826d2029a5c24946d199493e469d6ea78180ff
esh-pfi-infrastructure/services/intern-decision-serve/acceptance/systemone-2026-09-30/gpu1-32k-warm.csv
T
vh f65e27b08f feat(intern-decision): persist the triton autotune cache across recreates 
The ~6.5-9 s first-call-per-bucket autotune lived in the container's writable
layer and died on every recreate. 0.1.3 creates /tmp/triton-cache in the image
owned by 10001 so the named volume intern-decision_triton-cache inherits a
writable mount point, and compose mounts it.

Bucket model PROVEN, not inferred: 2,048-token buckets, 16 up to 32,768. After
one warmed call per bucket, 12 random sizes across 8k-32k were all warm (worst
2.09 s); cold entries cost 6.5-9 s. Full cold warm-up 109 s; warm re-run 17 s.

scripts/intern-decision-warmup: one noul call per bucket, MAX_TOKENS from
/health, two-point live calibration of the tokenizer's linear token model (a
single probe overcorrects and the aim oscillates around the bucket edge),
per-bucket wall times, non-zero exit on a missed bucket. Run it after an IMAGE
CHANGE only; the volume carries ordinary recreates (measured: force-recreate,
then a warmed 32k call answered in 2.11 s).

Acceptance on 0.1.3: JevBench 202/231, hard 83/111, 0 diffs / 924; warm 32k GPU
1 peak 15,218 MiB (budget 15,220; a COLD autotune touched 15,224 once, README
caveat); /decide answers. Artifacts in the acceptance dir.
2026-09-30 16:02:16 -07:00

3.0 KiB
Raw Blame History

18812
28812
38812
48812
58812
68812
78812
88812
98812
108812
118812
128812
138812
148812
158812
168812
178812
188812
198812
208812
218812
228812
238812
248812
258812
268812
278812
288812
298812
308812
318812
328812
338812
348812
358812
368812
378812
388812
398812
408812
418812
428812
438812
448812
458812
468812
478812
488812
498812
508812
518812
528812
538812
548812
558812
568812
578812
588812
598812
608812
618812
628812
638812
648812
658812
668812
678812
688812
698812
708812
718812
728812
738812
748812
758812
768812
778812
788812
798812
808812
818812
828812
838812
848812
858812
868812
878812
888812
898812
908812
918812
928812
9315218
9415218
9515218
9615218
9715218
9815218
9915218
10015218
10115218
10215218
10315218
10415218
10515218
10615218
10715218
10815218
10915218
11015218
11115218
11215218
11315218
11415218
11515218
11615218
1178812
1188812
1198812
1208812
1218812
1228812
1238812
1248812
1258812
1268812
1278812
1288812
1298812
1308812
1318812
1328812
1338812
1348812
1358812
1368812
1378812
1388812
1398812
1408812
1418812
1428812
1438812
1448812
1458812
1468812
1478812
1488812
1498812
1508812
1518812
1528812
1538812
1548812
1558812
1568812
1578812
1588812
1598812
1608812
1618812
1628812
1638812
1648812
1658812
1668812
1678812
1688812
1698812
1708812
1718812
1728812
1738812
1748812
1758812
1768812
1778812
1788812
1798812
1808812
1818812
1828812
1838812
1848812
1858812
1868812
1878812
1888812
1898812
1908812
1918812
1928812
1938812
1948812
1958812
1968812
1978812
1988812
1998812
2008812
2018812
2028812
2038812
2048812
2058812
2068812
2078812
2088812
2098812
2108812
2118812
2128812
2138812
2148812
2158812
2168812
2178812
2188812
2198812
2208812
2218812
2228832
2238832
2248832
2258834
2268834
2278834
2288834
2298834
2308834
2318834
2328834
2338834
2348834
2358834
2368834
2378834
2388834
2398834
2408834
2418834
2428834
2438834
2448834
2458834
2468834
2478834
2488834
2498834
2508834
2518834
2528812
2539306
2549300
2558812
2569300
2579300
2588884
2599118
2609118
2619200
2629300
2638812
2649296
2659298
2668812
2678988
2689646
2699540
2709500
2718812
2729566
2739518
2749032
2759122
2769124
2779124
2789380
2799352
2809346
2818812
2829352
2838916
2848916
2858916
2868916
2879350
2888812
2899352
2908880
2918880
2928944
2938944
2948944
2958944
2968944
2978944
2988944
2998944
3008944
3018944
3028944
3038944
3048944
3058944
3068944
3078944
3088944
3098944
3108944
3118944
3128944
3138944
3148944
3158944
3168944
3178944
3188944
3198944
3208944
3218944
3228944
3238944
3248944
3258944
3268944
3278944
3288944
3298944
3308944
3318944
3328944
3338944
3348944
3358944
3368944
3378944
3388944
3398944
3408944
3418944
3428944
3438944
3448944
3458944
3468944
3478944
3488944
3498944
3508944
3518944
3528944
3538944
3548944
3558944
3568944
3578944
3588944
3598944
3608944
3618944
3628944
3638944
3648944
3658944
3668944
3678944
3688944
3698944
3708944
3718944
3728944
3738944
3748944
3758944
3768944
3778944
3788944
3798944
3808944
3818944
3828944
3838944
3848944
3858944
3868944
3878944
3888944
3898944
3908944
3918944
3928944
3938944
3948944
3958944
3968944
3978944
3988944
3998944
4008944
4018944
4028944
4038944
4048944
4058944
4068944
4078944
4088944
4098944
4108944
4118944
4128944
4138944
4148944
4158944
4168944
4178944
4188944
4198944
4208944
4218944
4228944
4238944
4248944
4258944
4268944
4278944
4288944
4298944
4308944
4318944
4328944
4338944
4348944
4358944
4368944
4378944
4388944
4398944
4408944
4418944
4428944
4438944
4448944
4458944
4468944
4478944
4488944
4498944
4508944
4518944
4528944
4538944
4548944
4558944
4568944
4578944
4588944
4598944
4608944
4618944
4628944
4638944
4648944
4658944
4668944
4678944
4688944
4698944
4708944
4718944
4728944
4738944
4748944
4758944
4768944
4778944
4788944
4798944
4808944
4818944
4828944
4838944
4848944
4858944
4868944
4878944
4888944
4898944
4908944
4918944
4928944
4938944
4948944
4958944
4968944
4978944
4988944
4998944
5008944
5018944
5028944
5038944
5048944
5058944
5068944
5078944
5088944
5098944
5108944
5118944
5128944
5138944
5148944
5158944
5168944
5178944
5188944
5198944
5208944
5218944
5228944
5238944
5248944
5258944
5268944
5278944
5288944
5298944
5308944
5318944
5328944
5338944
5348944
5358944
5368944
5378944
5388944
5398944
5408944
5418944
5428944
5438944
5448944
5458944
5468944
5478944
5488944
5498944
5508944
5518944
5528944
5539294
5549346
5559346
5568812
5578812
5588812
5598812
5608812
5618812
5628812
5638812
5648812
5658812
5668812
5678812
5688812
5698812
5708812
5718812
5728812
5738812
5748812
5758812
5768812
5778812
5788812
5798812
5808812
5818812
5828812
5838812
5848812
5858812
5868812
5878812
5888812
5898812
5908812
5918812
5928812
5938812
5948812
5958812
5968812
5978812
5988812
5998812
6008812
Reference in New Issue View Git Blame Copy Permalink