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": 76728.67009537948,
-      "required_pages_per_second_upper_estimate": 0.005308267432241416
+      "required_token_positions_per_second": 76271.14535359986,
+      "required_pages_per_second_upper_estimate": 0.00527661480900144
     },
     {
       "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": 9591.083761922435,
-      "required_pages_per_second_upper_estimate": 0.004015586486512853
+      "required_token_positions_per_second": 9533.893169199982,
+      "required_pages_per_second_upper_estimate": 0.0039916419796153435
     },
     {
       "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": 19182.16752384487,
-      "required_pages_per_second_upper_estimate": 0.003995139627441148
+      "required_token_positions_per_second": 19067.786338399965,
+      "required_pages_per_second_upper_estimate": 0.003971317042947681
     },
     {
       "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": 28773.251285767306,
-      "required_pages_per_second_upper_estimate": 0.00172491248037091
+      "required_token_positions_per_second": 28601.679507599943,
+      "required_pages_per_second_upper_estimate": 0.001714627014244713
     },
     {
       "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": 28773.251285767306,
-      "required_pages_per_second_upper_estimate": 0.0022369937273112915
+      "required_token_positions_per_second": 28601.679507599943,
+      "required_pages_per_second_upper_estimate": 0.002223654776223276
     },
     {
       "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": 28773.251285767306,
-      "required_pages_per_second_upper_estimate": 0.3690723731836085
+      "required_token_positions_per_second": 28601.679507599943,
+      "required_pages_per_second_upper_estimate": 0.3668716346327002
     }
   ],
   "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": 0.0692518104140376,
-    "producer_p95_seconds": 0.16314581500046188,
-    "gpu_update_seconds": 0.9608924526953299,
-    "producer_batches_per_second": 14.440055704266413,
-    "gpu_batches_per_second": 1.040699192916931,
-    "producer_headroom": 13.875340542729743,
+    "producer_batch_seconds": 0.09592061485156478,
+    "producer_p95_seconds": 0.16391107399977045,
+    "gpu_update_seconds": 0.9666565207352059,
+    "producer_batches_per_second": 10.42528763548357,
+    "gpu_batches_per_second": 1.0344936164496508,
+    "producer_headroom": 10.07767227338031,
     "sustainable": true,
-    "buffer_seconds": 60.53622451980578,
+    "buffer_seconds": 60.89936080631797,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303738.50991712825,
-    "upload_bytes_per_second": 6121490.07474856,
+    "generation_bytes_per_second": 322278.4992899572,
+    "upload_bytes_per_second": 5932637.544627403,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.1752135575808,
+    "estimated_full_shards_per_hour": 69.15346368812597,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 11797840,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1542902260007395,
+      "bytes": 2688701,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 603500, "loss": 0.6747608184814453, "lr": 0.0003, "grad_norm": 0.2245971242190504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88334247204, "tokens_per_second": 147668.0634139631, "data_wait_s": 1.7825999748311006e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 603600, "loss": 0.6014395356178284, "lr": 0.0003, "grad_norm": 0.1580584142154489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88349061194, "tokens_per_second": 154145.45711414694, "data_wait_s": 1.6590000086580403e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 603700, "loss": 0.6376320123672485, "lr": 0.0003, "grad_norm": 0.2104245913898689, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88363819478, "tokens_per_second": 153228.70100129818, "data_wait_s": 1.696799972705776e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 603800, "loss": 0.5839520692825317, "lr": 0.0003, "grad_norm": 0.16259424906493905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88378426504, "tokens_per_second": 151527.46729536954, "data_wait_s": 1.4981000276748091e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0d71eb43129ff58597c87cdd4a4e57caadd84ed3f4a3d9821f4bedce7149ce2b
-size 78531
+oid sha256:9923a837c467dd990468e3c025286e92c193627e2fc7774259fa37681cdcd3b0
+size 80649