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": 76740.00056676529,
-      "required_pages_per_second_upper_estimate": 0.005309051300542202
+      "required_token_positions_per_second": 77060.41858217868,
+      "required_pages_per_second_upper_estimate": 0.005331218562320727
     },
     {
       "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": 9592.500070845661,
-      "required_pages_per_second_upper_estimate": 0.004016179465483115
+      "required_token_positions_per_second": 9632.552322772335,
+      "required_pages_per_second_upper_estimate": 0.004032948506978684
     },
     {
       "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": 19185.000141691322,
-      "required_pages_per_second_upper_estimate": 0.003995729587037408
+      "required_token_positions_per_second": 19265.10464554467,
+      "required_pages_per_second_upper_estimate": 0.004012413242692235
     },
     {
       "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": 28777.50021253698,
-      "required_pages_per_second_upper_estimate": 0.0017251671970430163
+      "required_token_positions_per_second": 28897.656968317006,
+      "required_pages_per_second_upper_estimate": 0.0017323704110832861
     },
     {
       "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": 28777.50021253698,
-      "required_pages_per_second_upper_estimate": 0.00223732406267858
+      "required_token_positions_per_second": 28897.656968317006,
+      "required_pages_per_second_upper_estimate": 0.0022466657219267633
     },
     {
       "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": 28777.50021253698,
-      "required_pages_per_second_upper_estimate": 0.3691268738540678
+      "required_token_positions_per_second": 28897.656968317006,
+      "required_pages_per_second_upper_estimate": 0.3706681157029413
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.2802722553122976,
-    "producer_p95_seconds": 0.23344139400069253,
-    "gpu_update_seconds": 0.960750579299973,
-    "producer_batches_per_second": 3.5679593004513945,
-    "gpu_batches_per_second": 1.0408528722705794,
-    "producer_headroom": 3.4279189648274038,
+    "producer_batch_seconds": 0.06090548632005266,
+    "producer_p95_seconds": 0.2087391530003515,
+    "gpu_update_seconds": 0.9567557684802227,
+    "producer_batches_per_second": 16.418882114249826,
+    "gpu_batches_per_second": 1.0451988197452622,
+    "producer_headroom": 15.708860174805276,
     "sustainable": true,
-    "buffer_seconds": 60.527286495898295,
+    "buffer_seconds": 60.27561341425403,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409051.82952523447,
-    "upload_bytes_per_second": 1894663.0423452999,
+    "generation_bytes_per_second": 410606.66852514265,
+    "upload_bytes_per_second": 1872463.712905144,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.77300037687088,
+    "estimated_full_shards_per_hour": 88.1066326314517,
     "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": 1.4379317339989939,
-      "bytes": 11505212,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.3395487400011916,
       "bytes": 6734300,
@@ -562,6 +556,12 @@
       "bytes": 7522866,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1316358820004098,
+      "bytes": 5807643,
+      "pages": 41,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -174,3 +174,4 @@
 {"step": 662400, "loss": 0.5223916172981262, "lr": 0.0001, "grad_norm": 0.19000770128861316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96958150355, "tokens_per_second": 151504.84606046116, "data_wait_s": 7.750000804662704e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 662500, "loss": 0.5135591626167297, "lr": 0.0001, "grad_norm": 0.18365159722230331, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96972774646, "tokens_per_second": 152532.11493162272, "data_wait_s": 8.673003321746364e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 662600, "loss": 0.48360851407051086, "lr": 0.0001, "grad_norm": 0.18846363952215442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96987475711, "tokens_per_second": 153696.57958388873, "data_wait_s": 1.367199729429558e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 662700, "loss": 0.5400761365890503, "lr": 0.0001, "grad_norm": 0.20251129177201113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97002009539, "tokens_per_second": 151179.20269854466, "data_wait_s": 8.656999852973968e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e342454766b61d1c60e9c8c6ed86fde162a10013f7d3c0fb4b489e9c2ba6bb90
-size 375282
+oid sha256:05ab7c06c1d91d9fecffdd39e46eedda1dffea3c77666229d9beb3287024f9f3
+size 377400