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": 76166.63031477894,
-      "required_pages_per_second_upper_estimate": 0.005269384216107508
+      "required_token_positions_per_second": 75362.74816648621,
+      "required_pages_per_second_upper_estimate": 0.00521376978382505
     },
     {
       "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": 9520.828789347368,
-      "required_pages_per_second_upper_estimate": 0.0039861721965863284
+      "required_token_positions_per_second": 9420.343520810777,
+      "required_pages_per_second_upper_estimate": 0.003944101113021078
     },
     {
       "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": 19041.657578694736,
-      "required_pages_per_second_upper_estimate": 0.003965875111362813
+      "required_token_positions_per_second": 18840.687041621553,
+      "required_pages_per_second_upper_estimate": 0.0039240182484399375
     },
     {
       "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": 28562.486368042104,
-      "required_pages_per_second_upper_estimate": 0.0017122774453726796
+      "required_token_positions_per_second": 28261.03056243233,
+      "required_pages_per_second_upper_estimate": 0.0016942056301227366
     },
     {
       "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": 28562.486368042104,
-      "required_pages_per_second_upper_estimate": 0.002220607682015056
+      "required_token_positions_per_second": 28261.03056243233,
+      "required_pages_per_second_upper_estimate": 0.0021971708190928526
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28562.486368042104,
-      "required_pages_per_second_upper_estimate": 0.36636890712076686
+      "required_token_positions_per_second": 28261.03056243233,
+      "required_pages_per_second_upper_estimate": 0.36250215572443056
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1038662158826753,
-    "producer_p95_seconds": 0.2904386770023848,
-    "gpu_update_seconds": 0.9679829565165132,
-    "producer_batches_per_second": 9.627769640992554,
-    "gpu_batches_per_second": 1.033076040510782,
-    "producer_headroom": 9.319516921747901,
+    "producer_batch_seconds": 0.09663364472672242,
+    "producer_p95_seconds": 0.27312077099850285,
+    "gpu_update_seconds": 0.9783082729032806,
+    "producer_batches_per_second": 10.348362651827689,
+    "gpu_batches_per_second": 1.0221726910601971,
+    "producer_headroom": 10.123888793286358,
     "sustainable": true,
-    "buffer_seconds": 60.01494330402382,
+    "buffer_seconds": 60.6551129200034,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450558.9683641795,
-    "upload_bytes_per_second": 5255600.4674839545,
+    "generation_bytes_per_second": 449642.85219412856,
+    "upload_bytes_per_second": 5222815.025768331,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.67946613496817,
+    "estimated_full_shards_per_hour": 96.48288893096822,
     "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": 2.0274646729994856,
-      "bytes": 10144494,
-      "pages": 44,
-      "shards": 3
-    },
     {
       "seconds": 2.5764055170002393,
       "bytes": 7638759,
@@ -562,6 +556,12 @@
       "bytes": 15475495,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.1073925720011175,
+      "bytes": 2592720,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -266,3 +266,4 @@
 {"step": 466800, "loss": 0.5992187261581421, "lr": 0.0003, "grad_norm": 0.16698154940396948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68325949648, "tokens_per_second": 152670.00531683842, "data_wait_s": 1.3724999007536098e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 466900, "loss": 0.6337512731552124, "lr": 0.0003, "grad_norm": 0.1660150600172574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68340525764, "tokens_per_second": 150606.48838295694, "data_wait_s": 1.109000004362315e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 467000, "loss": 0.6555135846138, "lr": 0.0003, "grad_norm": 0.23605228265184078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68355048751, "tokens_per_second": 150920.1984837939, "data_wait_s": 1.2935000995639712e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 467100, "loss": 0.5976812839508057, "lr": 0.0003, "grad_norm": 0.1841267185010909, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68369643579, "tokens_per_second": 150685.24245613866, "data_wait_s": 9.10999733605422e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5300c7b9b355feeffa00e521f1c50865ec8833d7b878a7e9b9526a3fb9dbde9c
-size 571704
+oid sha256:435b3e7264242b380bddf2a3a9d1d04bd3e3b3bbead63a97fca51a4eecf3dbac
+size 573822