Update TensorBoard training metrics

benchmark_cache/performance.json

@@ -49,11 +49,11 @@
   "selected": {
     "microbatch": 90,
     "mode": "compiled",
-    "microbatch_seconds": 0.9382242159999805,
+    "microbatch_seconds": 0.9440444560000287,
     "peak_gb": 89.37790203094482,
     "fits_budget": true,
-    "estimated_update_seconds": 0.9382242159999805,
-    "input_tokens_per_second": 196456.2381323185
+    "estimated_update_seconds": 0.9440444560000287,
+    "input_tokens_per_second": 195245.04257031978
   },
   "measurements": [
     {

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77570.75227723234,
-      "required_pages_per_second_upper_estimate": 0.005366524631481854
+      "required_token_positions_per_second": 78098.01702812791,
+      "required_pages_per_second_upper_estimate": 0.005403002030371586
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9696.344034654043,
-      "required_pages_per_second_upper_estimate": 0.00405965676462113
+      "required_token_positions_per_second": 9762.25212851599,
+      "required_pages_per_second_upper_estimate": 0.004087251107203621
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19392.688069308086,
-      "required_pages_per_second_upper_estimate": 0.004038985505261009
+      "required_token_positions_per_second": 19524.50425703198,
+      "required_pages_per_second_upper_estimate": 0.00406643934093726
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29089.032103962127,
-      "required_pages_per_second_upper_estimate": 0.0017438430582522985
+      "required_token_positions_per_second": 29286.756385547968,
+      "required_pages_per_second_upper_estimate": 0.0017556963270258727
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29089.032103962127,
-      "required_pages_per_second_upper_estimate": 0.002261544296952911
+      "required_token_positions_per_second": 29286.756385547968,
+      "required_pages_per_second_upper_estimate": 0.0022769164901490077
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29089.032103962127,
-      "required_pages_per_second_upper_estimate": 0.37312287046038567
+      "required_token_positions_per_second": 29286.756385547968,
+      "required_pages_per_second_upper_estimate": 0.3756590652447758
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,412 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12289854543001866,
-    "producer_p95_seconds": 0.35518173000309616,
-    "gpu_update_seconds": 0.9504613251203415,
-    "producer_batches_per_second": 8.136792803373117,
-    "gpu_batches_per_second": 1.0521206634824265,
-    "producer_headroom": 7.733706870123672,
-    "sustainable": true,
-    "buffer_seconds": 60.82952480770186,
-    "buffer_drain_seconds": null,
+    "producer_batch_seconds": 1.7146579959999997,
+    "producer_p95_seconds": 3.415684295999995,
+    "gpu_update_seconds": 0.9440444560000287,
+    "producer_batches_per_second": 0.5832066816431188,
+    "gpu_batches_per_second": 1.059272149361544,
+    "producer_headroom": 0.5505730345073601,
+    "sustainable": false,
+    "buffer_seconds": 60.41884518400184,
+    "buffer_drain_seconds": 134.43529165583922,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
-    "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416989.21227394184,
-    "upload_bytes_per_second": 4966327.20483757,
-    "upload_sustainable": true,
+    "upload_pages_per_commit": 8,
+    "max_pending_pages": 1806,
+    "generation_bytes_per_second": 0,
+    "upload_bytes_per_second": null,
+    "upload_sustainable": null,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.47617794192973,
+    "estimated_full_shards_per_hour": 0.0,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
-  "upload_measurements": [
-    {
-      "seconds": 1.5148740549993818,
-      "bytes": 14069210,
-      "pages": 39,
-      "shards": 4
-    },
-    {
-      "seconds": 1.4405004430009285,
-      "bytes": 5051876,
-      "pages": 30,
-      "shards": 2
-    },
-    {
-      "seconds": 1.163049670001783,
-      "bytes": 6248444,
-      "pages": 39,
-      "shards": 2
-    },
-    {
-      "seconds": 3.389049104996957,
-      "bytes": 2610225,
-      "pages": 33,
-      "shards": 2
-    },
-    {
-      "seconds": 1.3562926899976446,
-      "bytes": 6176571,
-      "pages": 34,
-      "shards": 2
-    },
-    {
-      "seconds": 1.3079960410032072,
-      "bytes": 12000675,
-      "pages": 34,
-      "shards": 4
-    },
-    {
-      "seconds": 1.1082588089993806,
-      "bytes": 7050197,
-      "pages": 37,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0682790589999058,
-      "bytes": 6087723,
-      "pages": 38,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1598256950019277,
-      "bytes": 7463538,
-      "pages": 29,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2493024350042106,
-      "bytes": 5521564,
-      "pages": 41,
-      "shards": 2
-    },
-    {
-      "seconds": 1.7268630670005223,
-      "bytes": 7581932,
-      "pages": 31,
-      "shards": 3
-    },
-    {
-      "seconds": 0.9802855540037854,
-      "bytes": 1121294,
-      "pages": 35,
-      "shards": 1
-    },
-    {
-      "seconds": 2.06611223800428,
-      "bytes": 11832847,
-      "pages": 31,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2940340749992174,
-      "bytes": 12187860,
-      "pages": 36,
-      "shards": 4
-    },
-    {
-      "seconds": 1.0110984060011106,
-      "bytes": 1179043,
-      "pages": 36,
-      "shards": 1
-    },
-    {
-      "seconds": 1.417530385995633,
-      "bytes": 5826314,
-      "pages": 31,
-      "shards": 2
-    },
-    {
-      "seconds": 1.3201328729992383,
-      "bytes": 7683555,
-      "pages": 38,
-      "shards": 3
-    },
-    {
-      "seconds": 1.031134179997025,
-      "bytes": 5438180,
-      "pages": 33,
-      "shards": 2
-    },
-    {
-      "seconds": 1.582061741995858,
-      "bytes": 13195042,
-      "pages": 34,
-      "shards": 4
-    },
-    {
-      "seconds": 0.8400354289988172,
-      "bytes": 1240516,
-      "pages": 36,
-      "shards": 1
-    },
-    {
-      "seconds": 1.1406454610041692,
-      "bytes": 5969910,
-      "pages": 37,
-      "shards": 2
-    },
-    {
-      "seconds": 1.6059057809980004,
-      "bytes": 11827384,
-      "pages": 24,
-      "shards": 4
-    },
-    {
-      "seconds": 1.091920383994875,
-      "bytes": 1379660,
-      "pages": 39,
-      "shards": 1
-    },
-    {
-      "seconds": 1.415137545001926,
-      "bytes": 7393395,
-      "pages": 35,
-      "shards": 3
-    },
-    {
-      "seconds": 1.3102802950015757,
-      "bytes": 11967731,
-      "pages": 38,
-      "shards": 3
-    },
-    {
-      "seconds": 0.9184283650029101,
-      "bytes": 675909,
-      "pages": 20,
-      "shards": 1
-    },
-    {
-      "seconds": 1.4080826390054426,
-      "bytes": 11404109,
-      "pages": 41,
-      "shards": 5
-    },
-    {
-      "seconds": 1.4213992930017412,
-      "bytes": 1225275,
-      "pages": 36,
-      "shards": 1
-    },
-    {
-      "seconds": 1.2654285969983903,
-      "bytes": 6147750,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 1.2639681400032714,
-      "bytes": 11134689,
-      "pages": 41,
-      "shards": 4
-    },
-    {
-      "seconds": 1.0753672879945952,
-      "bytes": 6680589,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 1.2491716029981035,
-      "bytes": 7586487,
-      "pages": 41,
-      "shards": 3
-    },
-    {
-      "seconds": 0.84151389899489,
-      "bytes": 763221,
-      "pages": 25,
-      "shards": 1
-    },
-    {
-      "seconds": 1.1390506280004047,
-      "bytes": 10406852,
-      "pages": 28,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2776995899985195,
-      "bytes": 7471801,
-      "pages": 42,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1862390570022399,
-      "bytes": 886866,
-      "pages": 26,
-      "shards": 1
-    },
-    {
-      "seconds": 1.5472574629966402,
-      "bytes": 11768607,
-      "pages": 31,
-      "shards": 3
-    },
-    {
-      "seconds": 1.4715010020008776,
-      "bytes": 7055090,
-      "pages": 43,
-      "shards": 3
-    },
-    {
-      "seconds": 2.0782521860019187,
-      "bytes": 6362598,
-      "pages": 36,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1697270270014997,
-      "bytes": 2277832,
-      "pages": 26,
-      "shards": 2
-    },
-    {
-      "seconds": 1.5861221969971666,
-      "bytes": 6049937,
-      "pages": 32,
-      "shards": 2
-    },
-    {
-      "seconds": 1.4160137470025802,
-      "bytes": 10237474,
-      "pages": 39,
-      "shards": 3
-    },
-    {
-      "seconds": 1.7253028289997019,
-      "bytes": 8025148,
-      "pages": 23,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0750127709980006,
-      "bytes": 6356822,
-      "pages": 43,
-      "shards": 2
-    },
-    {
-      "seconds": 1.7031787460000487,
-      "bytes": 7399641,
-      "pages": 32,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1420902980025858,
-      "bytes": 5869955,
-      "pages": 40,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1126535420044092,
-      "bytes": 5839811,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 1.9542204329991364,
-      "bytes": 7801020,
-      "pages": 41,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0194173400013824,
-      "bytes": 489708,
-      "pages": 17,
-      "shards": 1
-    },
-    {
-      "seconds": 1.3968911629999639,
-      "bytes": 16175519,
-      "pages": 45,
-      "shards": 4
-    },
-    {
-      "seconds": 1.1545158149965573,
-      "bytes": 2752375,
-      "pages": 38,
-      "shards": 2
-    },
-    {
-      "seconds": 1.3756929800001672,
-      "bytes": 5863415,
-      "pages": 30,
-      "shards": 2
-    },
-    {
-      "seconds": 2.5138282220068504,
-      "bytes": 7584714,
-      "pages": 39,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2855447649999405,
-      "bytes": 5546811,
-      "pages": 33,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1589639669982716,
-      "bytes": 5680180,
-      "pages": 19,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1662622910007485,
-      "bytes": 8697344,
-      "pages": 49,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2496280530031072,
-      "bytes": 6028853,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 1.3323977969994303,
-      "bytes": 12061980,
-      "pages": 38,
-      "shards": 5
-    },
-    {
-      "seconds": 0.9952729320066283,
-      "bytes": 1097506,
-      "pages": 36,
-      "shards": 1
-    },
-    {
-      "seconds": 1.100044867998804,
-      "bytes": 5862761,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 1.5496072789974278,
-      "bytes": 8395653,
-      "pages": 37,
-      "shards": 3
-    },
-    {
-      "seconds": 1.5835377559997141,
-      "bytes": 9557603,
-      "pages": 33,
-      "shards": 3
-    },
-    {
-      "seconds": 1.387810526000976,
-      "bytes": 7581888,
-      "pages": 40,
-      "shards": 3
-    },
-    {
-      "seconds": 1.081230299998424,
-      "bytes": 1007698,
-      "pages": 31,
-      "shards": 1
-    }
-  ],
+  "upload_measurements": [],
   "cache_key": {
     "version": 1,
     "kind": "data",
@@ -606,7 +221,7 @@
   },
   "cache_reused": true,
   "resume_probe_seconds": [
-    2.975958035000019,
-    0.01771132999999736
+    3.415684295999995,
+    0.013631696000004467
   ]
 }

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b65fc44666df32f089915a1a00aec1b2141ee82423f64dea5713a3d8a396621e
+size 165