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": 76909.5193712525,
-      "required_pages_per_second_upper_estimate": 0.005320779004774445
+      "required_token_positions_per_second": 76812.75913507461,
+      "required_pages_per_second_upper_estimate": 0.005314084907120978
     },
     {
       "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": 9613.689921406563,
-      "required_pages_per_second_upper_estimate": 0.004025051213418567
+      "required_token_positions_per_second": 9601.594891884326,
+      "required_pages_per_second_upper_estimate": 0.004019987277130543
     },
     {
       "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": 19227.379842813127,
-      "required_pages_per_second_upper_estimate": 0.004004556161153204
+      "required_token_positions_per_second": 19203.189783768652,
+      "required_pages_per_second_upper_estimate": 0.0039995180097889965
     },
     {
       "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": 28841.06976421969,
-      "required_pages_per_second_upper_estimate": 0.0017289780946013098
+      "required_token_positions_per_second": 28804.78467565298,
+      "required_pages_per_second_upper_estimate": 0.0017268028589458577
     },
     {
       "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": 28841.06976421969,
-      "required_pages_per_second_upper_estimate": 0.0022422663157090033
+      "required_token_positions_per_second": 28804.78467565298,
+      "required_pages_per_second_upper_estimate": 0.002239445309674179
     },
     {
       "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": 28841.06976421969,
-      "required_pages_per_second_upper_estimate": 0.3699422758073869
+      "required_token_positions_per_second": 28804.78467565298,
+      "required_pages_per_second_upper_estimate": 0.3694768496511458
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07562092510934804,
-    "producer_p95_seconds": 0.21557034400029806,
-    "gpu_update_seconds": 0.9586329573079908,
-    "producer_batches_per_second": 13.223853034778372,
-    "gpu_batches_per_second": 1.0431521182081775,
-    "producer_headroom": 12.67682134173584,
+    "producer_batch_seconds": 0.061677112867187134,
+    "producer_p95_seconds": 0.20545292799943127,
+    "gpu_update_seconds": 0.959840537303834,
+    "producer_batches_per_second": 16.213469689370147,
+    "gpu_batches_per_second": 1.0418397235117542,
+    "producer_headroom": 15.562345458204467,
     "sustainable": true,
-    "buffer_seconds": 60.39387631040342,
+    "buffer_seconds": 60.46995385014154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343299.90020249883,
+    "generation_bytes_per_second": 369475.62202023115,
     "upload_bytes_per_second": 5347045.17713968,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.66416697078918,
+    "estimated_full_shards_per_hour": 79.28086753325654,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 571600, "loss": 0.548515260219574, "lr": 0.0003, "grad_norm": 0.16169941412466862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83665578855, "tokens_per_second": 152346.60124668793, "data_wait_s": 1.0597999789752066e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 571700, "loss": 0.5836662650108337, "lr": 0.0003, "grad_norm": 0.1532148577348149, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83680151099, "tokens_per_second": 152232.6332315191, "data_wait_s": 1.2501999663072638e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 571800, "loss": 0.5821009278297424, "lr": 0.0003, "grad_norm": 0.16956327297832322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83694621871, "tokens_per_second": 150847.4240168994, "data_wait_s": 1.1878000805154443e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 571900, "loss": 0.5540379881858826, "lr": 0.0003, "grad_norm": 0.17463754712443672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83709282993, "tokens_per_second": 153536.64720869836, "data_wait_s": 5.871999746887013e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:edff2a04a062d5df8e8ac34f70275136a6b2288d354c8c7fa60f24c904143417
-size 313860
+oid sha256:81dce5c1c83a88ac09f37a9f864d56a093ccce646486a3348b51b96135cc91e0
+size 315978