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": 77094.34759828896,
-      "required_pages_per_second_upper_estimate": 0.0053335658503814
+      "required_token_positions_per_second": 77404.13185728864,
+      "required_pages_per_second_upper_estimate": 0.005354997444217509
     },
     {
       "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": 9636.79344978612,
-      "required_pages_per_second_upper_estimate": 0.004034724178294553
+      "required_token_positions_per_second": 9675.51648216108,
+      "required_pages_per_second_upper_estimate": 0.00405093670332108
     },
     {
       "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": 19273.58689957224,
-      "required_pages_per_second_upper_estimate": 0.004014179872514096
+      "required_token_positions_per_second": 19351.03296432216,
+      "required_pages_per_second_upper_estimate": 0.004030309845411432
     },
     {
       "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": 28910.380349358362,
-      "required_pages_per_second_upper_estimate": 0.0017331331583492374
+      "required_token_positions_per_second": 29026.549446483237,
+      "required_pages_per_second_upper_estimate": 0.001740097318341933
     },
     {
       "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": 28910.380349358362,
-      "required_pages_per_second_upper_estimate": 0.00224765490883849
+      "required_token_positions_per_second": 29026.549446483237,
+      "required_pages_per_second_upper_estimate": 0.0022566865451660923
     },
     {
       "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": 28910.380349358362,
-      "required_pages_per_second_upper_estimate": 0.37083131757363763
+      "required_token_positions_per_second": 29026.549446483237,
+      "required_pages_per_second_upper_estimate": 0.3723214100188971
     }
   ],
   "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": 0.1265192430000015,
-    "producer_p95_seconds": 0.1264875879999181,
-    "gpu_update_seconds": 0.9563347028263369,
-    "producer_batches_per_second": 7.90393600442257,
-    "gpu_batches_per_second": 1.0456590114785287,
-    "producer_headroom": 7.558808289947843,
+    "producer_batch_seconds": 0.06452713461718851,
+    "producer_p95_seconds": 0.12451029299995753,
+    "gpu_update_seconds": 0.9525072916770595,
+    "producer_batches_per_second": 15.49735635919007,
+    "gpu_batches_per_second": 1.0498607294011588,
+    "producer_headroom": 14.761344933846388,
     "sustainable": true,
-    "buffer_seconds": 60.249086278059224,
+    "buffer_seconds": 60.00795937565475,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 45,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 97,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472430.51055629563,
-    "upload_bytes_per_second": 22272.904219528333,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 654759.9919000895,
+    "upload_bytes_per_second": 3312805.9111200627,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.37258994595196,
+    "estimated_full_shards_per_hour": 70.24812611461647,
     "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": [
@@ -184,6 +184,12 @@
       "bytes": 21104,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.029576297999938,
+      "bytes": 6528629,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790911926322536473-f9bff804/train.jsonl

@@ -2,3 +2,4 @@
 {"step": 483400, "loss": 0.5242839455604553, "lr": 0.0003, "grad_norm": 0.15922517566295763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70753143675, "tokens_per_second": 154362.54684854476, "data_wait_s": 1.4137999983176996e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 483500, "loss": 0.5839016437530518, "lr": 0.0003, "grad_norm": 0.1528308329657215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70767836907, "tokens_per_second": 152693.96277359722, "data_wait_s": 2.2276999970927136e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 483600, "loss": 0.5361254215240479, "lr": 0.0003, "grad_norm": 0.15295186592155177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70782533160, "tokens_per_second": 152945.3604609795, "data_wait_s": 1.0869999982787704e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 483700, "loss": 0.6115140318870544, "lr": 0.0003, "grad_norm": 0.16111500090974964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70797232939, "tokens_per_second": 150965.23755505952, "data_wait_s": 1.635600006011373e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790911926.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.444.0.1790911926322536473-f9bff804

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d5b7240c77ae7e66d32846ab449ad1184f8e17cb4578a980d43b6b5307b49b70
-size 8637
+oid sha256:b330327f89e768c5ab92571c8304f12582399bf984b5088f4bed52627b7aed16
+size 10755