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": 76800.98689508256,
-      "required_pages_per_second_upper_estimate": 0.005313270476242969
+      "required_token_positions_per_second": 77479.97682908521,
+      "required_pages_per_second_upper_estimate": 0.005360244575350976
     },
     {
       "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": 9600.12336188532,
-      "required_pages_per_second_upper_estimate": 0.0040193711782490005
+      "required_token_positions_per_second": 9684.997103635651,
+      "required_pages_per_second_upper_estimate": 0.004054906041554594
     },
     {
       "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": 19200.24672377064,
-      "required_pages_per_second_upper_estimate": 0.003998905048005122
+      "required_token_positions_per_second": 19369.994207271302,
+      "required_pages_per_second_upper_estimate": 0.004034258972276135
     },
     {
       "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": 28800.370085655963,
-      "required_pages_per_second_upper_estimate": 0.001726538211016225
+      "required_token_positions_per_second": 29054.991310906953,
+      "required_pages_per_second_upper_estimate": 0.0017418023646859223
     },
     {
       "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": 28800.370085655963,
-      "required_pages_per_second_upper_estimate": 0.0022391020947196403
+      "required_token_positions_per_second": 29054.991310906953,
+      "required_pages_per_second_upper_estimate": 0.002258897775022496
     },
     {
       "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": 28800.370085655963,
-      "required_pages_per_second_upper_estimate": 0.36942022403068153
+      "required_token_positions_per_second": 29054.991310906953,
+      "required_pages_per_second_upper_estimate": 0.3726862317172298
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5151203785781036,
-    "producer_p95_seconds": 0.3522404249961255,
-    "gpu_update_seconds": 0.9599876639699884,
-    "producer_batches_per_second": 1.941293805460228,
-    "gpu_batches_per_second": 1.0416800522879037,
-    "producer_headroom": 1.8636181053831733,
+    "producer_batch_seconds": 0.11012314614094976,
+    "producer_p95_seconds": 0.35131151800305815,
+    "gpu_update_seconds": 0.9515748844716129,
+    "producer_batches_per_second": 9.080743104815326,
+    "gpu_batches_per_second": 1.050889442668799,
+    "producer_headroom": 8.64100707088104,
     "sustainable": true,
-    "buffer_seconds": 60.479222830109265,
+    "buffer_seconds": 60.900792606183224,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309583.1530064779,
-    "upload_bytes_per_second": 5024463.8753669895,
+    "generation_bytes_per_second": 296741.1419005673,
+    "upload_bytes_per_second": 5130079.894584893,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.42933790822748,
+    "estimated_full_shards_per_hour": 63.673741271617544,
     "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": 1.799824219000584,
-      "bytes": 6035033,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1105879839997215,
       "bytes": 2480881,
@@ -562,6 +556,12 @@
       "bytes": 637051,
       "pages": 18,
       "shards": 1
+    },
+    {
+      "seconds": 1.4846390739985509,
+      "bytes": 13365694,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -387,3 +387,4 @@
 {"step": 963700, "loss": 0.524150013923645, "lr": 3e-05, "grad_norm": 0.3474758651429756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141059586896, "tokens_per_second": 152881.51490051608, "data_wait_s": 1.1860996892210096e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 963800, "loss": 0.5416581630706787, "lr": 3e-05, "grad_norm": 0.33988183599747446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141074356592, "tokens_per_second": 129704.28218816551, "data_wait_s": 1.1614996765274554e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 963900, "loss": 0.4930093288421631, "lr": 3e-05, "grad_norm": 0.35177214582856164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141089051260, "tokens_per_second": 153172.66581548544, "data_wait_s": 1.8566002836450934e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 964000, "loss": 0.5064939260482788, "lr": 3e-05, "grad_norm": 0.3384078411223471, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141103593687, "tokens_per_second": 150458.0840729264, "data_wait_s": 1.1809999705292284e-05, "gpu_peak_gb": 89.37899398803711}

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

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a205e542e663c69b979d1d12979b38c0f8f3a110c9301785d06606a783c1d996
-size 829548
+oid sha256:f207f4bb49222ca4b0ed6e9f46940ed5ff280eda05a008667aac8b206c015dbb
+size 831666