Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75750.67792838323,
-      "required_pages_per_second_upper_estimate": 0.006565876774494292
+      "required_token_positions_per_second": 75020.40006333936,
+      "required_pages_per_second_upper_estimate": 0.006502578140024616
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9468.834741047904,
-      "required_pages_per_second_upper_estimate": 0.003777259749899435
+      "required_token_positions_per_second": 9377.55000791742,
+      "required_pages_per_second_upper_estimate": 0.003740844905025299
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18937.669482095807,
-      "required_pages_per_second_upper_estimate": 0.004380896543083302
+      "required_token_positions_per_second": 18755.10001583484,
+      "required_pages_per_second_upper_estimate": 0.004338662310176692
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28406.50422314371,
-      "required_pages_per_second_upper_estimate": 0.001729807069780011
+      "required_token_positions_per_second": 28132.65002375226,
+      "required_pages_per_second_upper_estimate": 0.001713130785839014
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28406.50422314371,
-      "required_pages_per_second_upper_estimate": 0.0021868113293341076
+      "required_token_positions_per_second": 28132.65002375226,
+      "required_pages_per_second_upper_estimate": 0.0021657292749880106
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28406.50422314371,
-      "required_pages_per_second_upper_estimate": 0.35271371199751306
+      "required_token_positions_per_second": 28132.65002375226,
+      "required_pages_per_second_upper_estimate": 0.3493133593721179
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08887273362506676,
-    "producer_p95_seconds": 0.34691206400020747,
-    "gpu_update_seconds": 0.9732982201123596,
-    "producer_batches_per_second": 11.252045022253572,
-    "gpu_batches_per_second": 1.0274343252005105,
-    "producer_headroom": 10.951595392783537,
+    "producer_batch_seconds": 0.10542255930468514,
+    "producer_p95_seconds": 0.3484501189996081,
+    "gpu_update_seconds": 0.9827726849997042,
+    "producer_batches_per_second": 9.485635774690953,
+    "gpu_batches_per_second": 1.017529297734095,
+    "producer_headroom": 9.322223739222277,
     "sustainable": true,
-    "buffer_seconds": 60.344489646966295,
+    "buffer_seconds": 60.931906469981655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430865.95374321163,
+    "generation_bytes_per_second": 433436.8845539453,
     "upload_bytes_per_second": 4543113.590942459,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.45380362722646,
+    "estimated_full_shards_per_hour": 93.00546553100365,
     "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": [

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -319,3 +319,4 @@
 {"step": 186900, "loss": 0.587534736841917, "lr": 0.0003, "grad_norm": 0.14738928761525955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27353807895, "tokens_per_second": 145889.4209428797, "data_wait_s": 1.0378000297350809e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 187000, "loss": 0.5876833312213421, "lr": 0.0003, "grad_norm": 0.13868859085156893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27368515947, "tokens_per_second": 151428.8942013001, "data_wait_s": 1.1518000974319875e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 187100, "loss": 0.6052984334528446, "lr": 0.0003, "grad_norm": 0.12999672994602687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27383160466, "tokens_per_second": 149658.03758488997, "data_wait_s": 1.3742999726673588e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 187200, "loss": 0.6308657303452492, "lr": 0.0003, "grad_norm": 0.1502174377761144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27397811976, "tokens_per_second": 150745.3218314106, "data_wait_s": 8.782000804785639e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:76cf25fec2c64f1256833094c244de790aeb306640463e05dce47fc0bd3002f2
-size 684741
+oid sha256:a934f506794cca22c3eca9dd047f308b3f9988ccae8fb748deccaeab16ddee6a
+size 686859