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": 77116.11227085207,
-      "required_pages_per_second_upper_estimate": 0.005335071580930314
+      "required_token_positions_per_second": 76450.93956396845,
+      "required_pages_per_second_upper_estimate": 0.005289053389654783
     },
     {
       "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": 9639.514033856509,
-      "required_pages_per_second_upper_estimate": 0.004035863230032568
+      "required_token_positions_per_second": 9556.367445496056,
+      "required_pages_per_second_upper_estimate": 0.004001051489784253
     },
     {
       "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": 19279.028067713018,
-      "required_pages_per_second_upper_estimate": 0.004015313124344566
+      "required_token_positions_per_second": 19112.734890992113,
+      "required_pages_per_second_upper_estimate": 0.0039806786410795806
     },
     {
       "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": 28918.542101569525,
-      "required_pages_per_second_upper_estimate": 0.0017336224429320226
+      "required_token_positions_per_second": 28669.10233648817,
+      "required_pages_per_second_upper_estimate": 0.0017186689098878634
     },
     {
       "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": 28918.542101569525,
-      "required_pages_per_second_upper_estimate": 0.002248289449173154
+      "required_token_positions_per_second": 28669.10233648817,
+      "required_pages_per_second_upper_estimate": 0.0022288966046076527
     },
     {
       "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": 28918.542101569525,
-      "required_pages_per_second_upper_estimate": 0.37093600776759567
+      "required_token_positions_per_second": 28669.10233648817,
+      "required_pages_per_second_upper_estimate": 0.36773646228868495
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.15695111967966113,
-    "producer_p95_seconds": 0.20728992399926938,
-    "gpu_update_seconds": 0.9560647940996801,
-    "producer_batches_per_second": 6.371410424092612,
-    "gpu_batches_per_second": 1.0459542137431108,
-    "producer_headroom": 6.091481195234659,
+    "producer_batch_seconds": 0.08913444406245219,
+    "producer_p95_seconds": 0.21124960199995257,
+    "gpu_update_seconds": 0.9643831772441449,
+    "producer_batches_per_second": 11.219007539883778,
+    "gpu_batches_per_second": 1.036932231499138,
+    "producer_headroom": 10.819422136839137,
     "sustainable": true,
-    "buffer_seconds": 60.23208202827985,
+    "buffer_seconds": 60.75614016638113,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394632.28411177377,
-    "upload_bytes_per_second": 5545390.910560467,
+    "generation_bytes_per_second": 431806.37453247455,
+    "upload_bytes_per_second": 5585314.898140948,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.6789016009799,
+    "estimated_full_shards_per_hour": 92.65559603672673,
     "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": 1.2178643509996618,
-      "bytes": 7619131,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 0.8663683109998601,
       "bytes": 939323,
@@ -562,6 +556,12 @@
       "bytes": 2353349,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.4940221720007685,
+      "bytes": 12237361,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -155,3 +155,4 @@
 {"step": 572600, "loss": 0.6073922514915466, "lr": 0.0003, "grad_norm": 0.17427609695454738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83811616236, "tokens_per_second": 151883.46306007996, "data_wait_s": 8.850000085658394e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 572700, "loss": 0.6181311011314392, "lr": 0.0003, "grad_norm": 0.18139924881501968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83826301624, "tokens_per_second": 153689.98038245426, "data_wait_s": 1.3373000911087729e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 572800, "loss": 0.5630581378936768, "lr": 0.0003, "grad_norm": 0.17188633637176376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83840924034, "tokens_per_second": 152890.5815956797, "data_wait_s": 8.622000677860342e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 572900, "loss": 0.5719613432884216, "lr": 0.0003, "grad_norm": 0.15580971464201138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83855622847, "tokens_per_second": 152768.45641343156, "data_wait_s": 9.494999176240526e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c3421180380b3e873a34c22cc5dc16b3dcfe7ae38db1eb7be0891772697d66a8
-size 335040
+oid sha256:78863d30cda3bfb5c445234263aabffb86abd8b1b6e8b32371cd65ce83d54b1f
+size 337158