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": 76098.06327116607,
-      "required_pages_per_second_upper_estimate": 0.006595987255566606
+      "required_token_positions_per_second": 75254.2070029077,
+      "required_pages_per_second_upper_estimate": 0.006522843932968129
     },
     {
       "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": 9512.25790889576,
-      "required_pages_per_second_upper_estimate": 0.0037945819007881598
+      "required_token_positions_per_second": 9406.775875363463,
+      "required_pages_per_second_upper_estimate": 0.0037525035405151837
     },
     {
       "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": 19024.51581779152,
-      "required_pages_per_second_upper_estimate": 0.004400986914403199
+      "required_token_positions_per_second": 18813.551750726925,
+      "required_pages_per_second_upper_estimate": 0.0043521840903285765
     },
     {
       "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": 28536.773726687275,
-      "required_pages_per_second_upper_estimate": 0.0017377397990745505
+      "required_token_positions_per_second": 28220.327626090388,
+      "required_pages_per_second_upper_estimate": 0.0017184698918125778
     },
     {
       "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": 28536.773726687275,
-      "required_pages_per_second_upper_estimate": 0.002196839836325957
+      "required_token_positions_per_second": 28220.327626090388,
+      "required_pages_per_second_upper_estimate": 0.002172478939523137
     },
     {
       "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": 28536.773726687275,
-      "required_pages_per_second_upper_estimate": 0.3543312232475418
+      "required_token_positions_per_second": 28220.327626090388,
+      "required_pages_per_second_upper_estimate": 0.35040202175509877
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.295743847218489,
-    "producer_p95_seconds": 0.3785572989945649,
-    "gpu_update_seconds": 0.968855143359948,
-    "producer_batches_per_second": 3.3813044951066122,
-    "gpu_batches_per_second": 1.0321460404617795,
-    "producer_headroom": 3.2759942513501534,
+    "producer_batch_seconds": 0.14101665164082533,
+    "producer_p95_seconds": 0.37860732700210065,
+    "gpu_update_seconds": 0.979719313196022,
+    "producer_batches_per_second": 7.091361114906042,
+    "gpu_batches_per_second": 1.0207005073094035,
+    "producer_headroom": 6.947543441120724,
     "sustainable": true,
-    "buffer_seconds": 60.069018888316776,
+    "buffer_seconds": 60.742597418153366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 513317.3724630384,
+    "generation_bytes_per_second": 521910.4035556761,
     "upload_bytes_per_second": 5586375.47026958,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.14595871370663,
+    "estimated_full_shards_per_hour": 111.98982315066063,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -314,3 +314,4 @@
 {"step": 311600, "loss": 0.5842957980930805, "lr": 0.0003, "grad_norm": 0.16250620971312138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45609751924, "tokens_per_second": 149430.65663408334, "data_wait_s": 1.084099494619295e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 311700, "loss": 0.5714015923440456, "lr": 0.0003, "grad_norm": 0.14300842317176646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45624203878, "tokens_per_second": 148448.59188599358, "data_wait_s": 9.22300387173891e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 311800, "loss": 0.6630593258887529, "lr": 0.0003, "grad_norm": 0.14035228841223935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45638820990, "tokens_per_second": 149819.8939472595, "data_wait_s": 1.1156000255141407e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 311900, "loss": 0.5738855935633183, "lr": 0.0003, "grad_norm": 0.14206522713526085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45653498694, "tokens_per_second": 149941.06140293326, "data_wait_s": 7.285998435690999e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:964ff9d89db2b017838e0113e2e94a59f6ba19e4a4659d152a0d4c1dd9ed7324
-size 674151
+oid sha256:2a3a4014fbb98b0835e6294ef8763035faf4d02b568105b5d945656618531689
+size 676269