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": 76732.07735116074,
-      "required_pages_per_second_upper_estimate": 0.0053085031541022865
+      "required_token_positions_per_second": 76649.71717301355,
+      "required_pages_per_second_upper_estimate": 0.005302805285876121
     },
     {
       "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": 9591.509668895093,
-      "required_pages_per_second_upper_estimate": 0.004015764804868356
+      "required_token_positions_per_second": 9581.214646626693,
+      "required_pages_per_second_upper_estimate": 0.004011454494029089
     },
     {
       "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": 19183.019337790185,
-      "required_pages_per_second_upper_estimate": 0.003995317037822113
+      "required_token_positions_per_second": 19162.429293253386,
+      "required_pages_per_second_upper_estimate": 0.003991028674541087
     },
     {
       "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": 28774.529006685276,
-      "required_pages_per_second_upper_estimate": 0.0017249890777889755
+      "required_token_positions_per_second": 28743.64393988008,
+      "required_pages_per_second_upper_estimate": 0.0017231375651927702
     },
     {
       "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": 28774.529006685276,
-      "required_pages_per_second_upper_estimate": 0.0022370930644925635
+      "required_token_positions_per_second": 28743.64393988008,
+      "required_pages_per_second_upper_estimate": 0.0022346918863975114
     },
     {
       "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": 28774.529006685276,
-      "required_pages_per_second_upper_estimate": 0.36908876241563443
+      "required_token_positions_per_second": 28743.64393988008,
+      "required_pages_per_second_upper_estimate": 0.36869260194045844
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10379790004657252,
-    "producer_p95_seconds": 0.30956860299920663,
-    "gpu_update_seconds": 0.9608497846681158,
-    "producer_batches_per_second": 9.634106273357316,
-    "gpu_batches_per_second": 1.0407454067811515,
-    "producer_headroom": 9.256928938225121,
+    "producer_batch_seconds": 0.11052775129718384,
+    "producer_p95_seconds": 0.33612018099665875,
+    "gpu_update_seconds": 0.9618822184768321,
+    "producer_batches_per_second": 9.047501539330414,
+    "gpu_batches_per_second": 1.03962832537182,
+    "producer_headroom": 8.702630852323692,
     "sustainable": true,
-    "buffer_seconds": 60.533536434091296,
+    "buffer_seconds": 60.59857976404042,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493037.7153887816,
+    "generation_bytes_per_second": 487685.6556864421,
     "upload_bytes_per_second": 5056775.41149406,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.7944163918265,
+    "estimated_full_shards_per_hour": 104.64598896927784,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 683200, "loss": 0.5725772380828857, "lr": 0.0001, "grad_norm": 0.2014802505878475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100002652287, "tokens_per_second": 153065.17170915345, "data_wait_s": 1.7001002561300993e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 683300, "loss": 0.5685223937034607, "lr": 0.0001, "grad_norm": 0.20739122870950957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100017359825, "tokens_per_second": 152272.69776061346, "data_wait_s": 1.486000110162422e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 683400, "loss": 0.5565347075462341, "lr": 0.0001, "grad_norm": 0.1834445118323151, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100032050677, "tokens_per_second": 152632.3169792421, "data_wait_s": 1.0711002687457949e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 683500, "loss": 0.49607446789741516, "lr": 0.0001, "grad_norm": 0.19096895211888984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100046773572, "tokens_per_second": 154159.41512214197, "data_wait_s": 1.0621995897963643e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7fdbfcafde01ecbbf4a4522a78a66f477a131dd9d14a25e35d52871a69303359
-size 818958
+oid sha256:a7ac798f78e51ff52749022b20c4a0b3804c649c25bea57cce2f5a6f4801177d
+size 821076