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": 75418.54154902343,
-      "required_pages_per_second_upper_estimate": 0.006537088034923342
+      "required_token_positions_per_second": 75825.86474517194,
+      "required_pages_per_second_upper_estimate": 0.006572393777214292
     },
     {
       "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": 9427.317693627929,
-      "required_pages_per_second_upper_estimate": 0.0037606979789484318
+      "required_token_positions_per_second": 9478.233093146493,
+      "required_pages_per_second_upper_estimate": 0.0037810088930694485
     },
     {
       "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": 18854.635387255857,
-      "required_pages_per_second_upper_estimate": 0.004361688066592237
+      "required_token_positions_per_second": 18956.466186292986,
+      "required_pages_per_second_upper_estimate": 0.004385244829788624
     },
     {
       "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": 28281.953080883788,
-      "required_pages_per_second_upper_estimate": 0.0017222225586857225
+      "required_token_positions_per_second": 28434.69927943948,
+      "required_pages_per_second_upper_estimate": 0.0017315240007803954
     },
     {
       "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": 28281.953080883788,
-      "required_pages_per_second_upper_estimate": 0.00217722303762331
+      "required_token_positions_per_second": 28434.69927943948,
+      "required_pages_per_second_upper_estimate": 0.002188981862816667
     },
     {
       "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": 28281.953080883788,
-      "required_pages_per_second_upper_estimate": 0.35116720365650306
+      "required_token_positions_per_second": 28434.69927943948,
+      "required_pages_per_second_upper_estimate": 0.35306380023392325
     }
   ],
   "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": 0.07674699639844107,
-    "producer_p95_seconds": 0.20386882699858688,
-    "gpu_update_seconds": 0.9775845367160203,
-    "producer_batches_per_second": 13.029825881502674,
-    "gpu_batches_per_second": 1.0229294372426139,
-    "producer_headroom": 12.737756297859203,
+    "producer_batch_seconds": 0.0806619277890519,
+    "producer_p95_seconds": 0.22527127800094604,
+    "gpu_update_seconds": 0.9723331246900745,
+    "producer_batches_per_second": 12.39742251902549,
+    "gpu_batches_per_second": 1.0284541116695414,
+    "producer_headroom": 12.054424576027149,
     "sustainable": true,
-    "buffer_seconds": 60.61024127639326,
+    "buffer_seconds": 60.28465373078462,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 484478.82504440047,
-    "upload_bytes_per_second": 6019847.24316661,
+    "generation_bytes_per_second": 469064.36228223785,
+    "upload_bytes_per_second": 6009279.239589546,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.95787776469241,
+    "estimated_full_shards_per_hour": 100.65029288626053,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 11838640,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0752307399998244,
+      "bytes": 5923485,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -103,3 +103,4 @@
 {"step": 125300, "loss": 0.5879206787794828, "lr": 0.0003, "grad_norm": 0.1335140020427302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18336475000, "tokens_per_second": 149459.29210966578, "data_wait_s": 1.6451000192319043e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 125400, "loss": 0.6304312646389008, "lr": 0.0003, "grad_norm": 0.15239185870695368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18351272583, "tokens_per_second": 145620.47833194915, "data_wait_s": 9.3830003606854e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 125500, "loss": 0.6334497854113579, "lr": 0.0003, "grad_norm": 0.1416959142518137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18365841451, "tokens_per_second": 148438.85682220917, "data_wait_s": 1.3680999472853728e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 125600, "loss": 0.6110418401658535, "lr": 0.0003, "grad_norm": 0.143462033952661, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18380539739, "tokens_per_second": 150167.48174788724, "data_wait_s": 9.536001016385853e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd8ea9499c45c9c01c0dfa7761db0f2c651e26bb1acb7bf91d325c9d28ed23b2
-size 224121
+oid sha256:629ac6b2f9174ef9eb59db557b25d8a16b4d73b95467166f969fba00958c7d7b
+size 226239