Update TensorBoard training metrics

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": 77554.4145090482,
-      "required_pages_per_second_upper_estimate": 0.00536539434677003
+      "required_token_positions_per_second": 77214.77852323925,
+      "required_pages_per_second_upper_estimate": 0.005341897541207705
     },
     {
       "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": 9694.301813631026,
-      "required_pages_per_second_upper_estimate": 0.004058801729325217
+      "required_token_positions_per_second": 9651.847315404906,
+      "required_pages_per_second_upper_estimate": 0.004041026917468677
     },
     {
       "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": 19388.60362726205,
-      "required_pages_per_second_upper_estimate": 0.0040381348236968005
+      "required_token_positions_per_second": 19303.694630809812,
+      "required_pages_per_second_upper_estimate": 0.004020450518936616
     },
     {
       "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": 29082.905440893075,
-      "required_pages_per_second_upper_estimate": 0.0017434757741561818
+      "required_token_positions_per_second": 28955.54194621472,
+      "required_pages_per_second_upper_estimate": 0.0017358405271230104
     },
     {
       "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": 29082.905440893075,
-      "required_pages_per_second_upper_estimate": 0.002261067975847635
+      "required_token_positions_per_second": 28955.54194621472,
+      "required_pages_per_second_upper_estimate": 0.002251166024349201
     },
     {
       "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": 29082.905440893075,
-      "required_pages_per_second_upper_estimate": 0.37304428420483415
+      "required_token_positions_per_second": 28955.54194621472,
+      "required_pages_per_second_upper_estimate": 0.3714106020473662
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10195158224189527,
-    "producer_p95_seconds": 0.2959083680034382,
-    "gpu_update_seconds": 0.9506615512054214,
-    "producer_batches_per_second": 9.80857754249808,
-    "gpu_batches_per_second": 1.0518990683193385,
-    "producer_headroom": 9.324637541669885,
+    "producer_batch_seconds": 0.14198820184338956,
+    "producer_p95_seconds": 0.3053699040028732,
+    "gpu_update_seconds": 0.9548431195436268,
+    "producer_batches_per_second": 7.0428386796741185,
+    "gpu_batches_per_second": 1.0472924604389005,
+    "producer_headroom": 6.724806055342553,
     "sustainable": true,
-    "buffer_seconds": 60.84233927714697,
+    "buffer_seconds": 60.15511653124849,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 425194.20985491655,
-    "upload_bytes_per_second": 5008563.59762069,
+    "generation_bytes_per_second": 445169.05986301013,
+    "upload_bytes_per_second": 4877066.911888643,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.2367794202387,
+    "estimated_full_shards_per_hour": 95.52291724126556,
     "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": 1.7035883720018319,
-      "bytes": 12069614,
-      "pages": 41,
-      "shards": 3
-    },
-    {
-      "seconds": 1.154275818997121,
-      "bytes": 1133519,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.6667711109985248,
       "bytes": 11320824,
@@ -562,6 +550,18 @@
       "bytes": 5875734,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 2.4478168570058187,
+      "bytes": 1176227,
+      "pages": 37,
+      "shards": 1
+    },
+    {
+      "seconds": 1.676982698001666,
+      "bytes": 6919911,
+      "pages": 22,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -348,3 +348,4 @@
 {"step": 919800, "loss": 0.5458102226257324, "lr": 3e-05, "grad_norm": 0.3303582873094365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134632073688, "tokens_per_second": 154591.87344219937, "data_wait_s": 9.939998562913388e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 919900, "loss": 0.5420734882354736, "lr": 3e-05, "grad_norm": 0.33643525625140286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134646730041, "tokens_per_second": 153581.4379652662, "data_wait_s": 1.695499668130651e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 920000, "loss": 0.5171075463294983, "lr": 3e-05, "grad_norm": 0.33507429154060336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134661204932, "tokens_per_second": 150984.5966808495, "data_wait_s": 1.747399801388383e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 920100, "loss": 0.5363869667053223, "lr": 3e-05, "grad_norm": 0.3422005954617982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134675705300, "tokens_per_second": 135160.30847152337, "data_wait_s": 1.5227997209876776e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:77fb602a919992df34e45a23cef1ff64df9b1b675be2d544f90fca08fe907e10
-size 746946
+oid sha256:c9e89acf80b7086f4f651d98c160ebb08dc8899c5e1ac2c615e3185473e62e90
+size 749064