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": 76189.62621202419,
-      "required_pages_per_second_upper_estimate": 0.005270975125636767
+      "required_token_positions_per_second": 76545.07630871964,
+      "required_pages_per_second_upper_estimate": 0.0052955659880840125
     },
     {
       "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": 9523.703276503024,
-      "required_pages_per_second_upper_estimate": 0.003987375684332284
+      "required_token_positions_per_second": 9568.134538589955,
+      "required_pages_per_second_upper_estimate": 0.004005978125937822
     },
     {
       "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": 19047.406553006047,
-      "required_pages_per_second_upper_estimate": 0.00396707247110121
+      "required_token_positions_per_second": 19136.26907717991,
+      "required_pages_per_second_upper_estimate": 0.003985580191424273
     },
     {
       "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": 28571.10982950907,
-      "required_pages_per_second_upper_estimate": 0.0017127944087203608
+      "required_token_positions_per_second": 28704.403615769865,
+      "required_pages_per_second_upper_estimate": 0.0017207851676788685
     },
     {
       "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": 28571.10982950907,
-      "required_pages_per_second_upper_estimate": 0.002221278118213514
+      "required_token_positions_per_second": 28704.403615769865,
+      "required_pages_per_second_upper_estimate": 0.002231641123797887
     },
     {
       "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": 28571.10982950907,
-      "required_pages_per_second_upper_estimate": 0.36647951962531355
+      "required_token_positions_per_second": 28704.403615769865,
+      "required_pages_per_second_upper_estimate": 0.36818926919575
     }
   ],
   "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": 0.4531342907499436,
-    "producer_p95_seconds": 0.29220848299883073,
-    "gpu_update_seconds": 0.9676907955267577,
-    "producer_batches_per_second": 2.2068513030540813,
-    "gpu_batches_per_second": 1.0333879423288872,
-    "producer_headroom": 2.135549693061666,
+    "producer_batch_seconds": 0.1277295459215395,
+    "producer_p95_seconds": 0.31835309300004155,
+    "gpu_update_seconds": 0.9631971585297285,
+    "producer_batches_per_second": 7.829042159237538,
+    "gpu_batches_per_second": 1.0382090428157504,
+    "producer_headroom": 7.540911161787047,
     "sustainable": true,
-    "buffer_seconds": 60.96452011818574,
+    "buffer_seconds": 60.681420987372896,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 327027.87622578803,
-    "upload_bytes_per_second": 5334191.644855798,
+    "generation_bytes_per_second": 342007.6797657405,
+    "upload_bytes_per_second": 5253240.9875396,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.17256941871862,
+    "estimated_full_shards_per_hour": 73.38688654641305,
     "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": 1.1967464800000016,
-      "bytes": 10837616,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.1927219149984012,
       "bytes": 6126779,
@@ -562,6 +556,12 @@
       "bytes": 7617654,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.338195986001665,
+      "bytes": 5044470,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -308,3 +308,4 @@
 {"step": 635200, "loss": 0.5750125050544739, "lr": 0.0001, "grad_norm": 0.1774653268777128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92976760316, "tokens_per_second": 153024.1355079562, "data_wait_s": 1.801799953682348e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 635300, "loss": 0.5325800180435181, "lr": 0.0001, "grad_norm": 0.17676632920750399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92991266546, "tokens_per_second": 150612.82307274116, "data_wait_s": 8.048002200666815e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 635400, "loss": 0.49283096194267273, "lr": 0.0001, "grad_norm": 0.18659582375256248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93005764105, "tokens_per_second": 149815.13955222492, "data_wait_s": 1.4457997167482972e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 635500, "loss": 0.5993461012840271, "lr": 0.0001, "grad_norm": 0.16563936719817438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93020257613, "tokens_per_second": 150486.24634771576, "data_wait_s": 1.4549001207342371e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50036c5b1f0ac3bd8c5a26103d68a5725d0d914b2f6481a0aeea9ed2f8b8be4d
-size 662226
+oid sha256:4f177c17aaedc84fd6f7d8f668b354f8334e043aae170b36a2ac9280809b9201
+size 664344