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": 77241.39166139596,
-      "required_pages_per_second_upper_estimate": 0.005343738699856375
+      "required_token_positions_per_second": 76872.30517150898,
+      "required_pages_per_second_upper_estimate": 0.0053182044400873485
     },
     {
       "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": 9655.173957674495,
-      "required_pages_per_second_upper_estimate": 0.004042419713118764
+      "required_token_positions_per_second": 9609.038146438623,
+      "required_pages_per_second_upper_estimate": 0.004023103612379598
     },
     {
       "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": 19310.34791534899,
-      "required_pages_per_second_upper_estimate": 0.004021836222647214
+      "required_token_positions_per_second": 19218.076292877246,
+      "required_pages_per_second_upper_estimate": 0.004002618477052671
     },
     {
       "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": 28965.52187302349,
-      "required_pages_per_second_upper_estimate": 0.0017364388084967302
+      "required_token_positions_per_second": 28827.11443931587,
+      "required_pages_per_second_upper_estimate": 0.0017281414941818728
     },
     {
       "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": 28965.52187302349,
-      "required_pages_per_second_upper_estimate": 0.002251941919761524
+      "required_token_positions_per_second": 28827.11443931587,
+      "required_pages_per_second_upper_estimate": 0.002241181350580718
     },
     {
       "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": 28965.52187302349,
-      "required_pages_per_second_upper_estimate": 0.3715386138328586
+      "required_token_positions_per_second": 28827.11443931587,
+      "required_pages_per_second_upper_estimate": 0.3697632718835811
     }
   ],
   "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": 0.057142534156326974,
-    "producer_p95_seconds": 0.20159809700089681,
-    "gpu_update_seconds": 0.9545141330855655,
-    "producer_batches_per_second": 17.50009891518396,
-    "gpu_batches_per_second": 1.0476534242268334,
-    "producer_headroom": 16.704091744938463,
+    "producer_batch_seconds": 0.1094828718124603,
+    "producer_p95_seconds": 0.20776308000131394,
+    "gpu_update_seconds": 0.9590970354733899,
+    "producer_batches_per_second": 9.133848824434924,
+    "gpu_batches_per_second": 1.0426473683201631,
+    "producer_headroom": 8.760247329977643,
     "sustainable": true,
-    "buffer_seconds": 60.13439038439063,
+    "buffer_seconds": 60.42311323482356,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 502051.06093470246,
+    "generation_bytes_per_second": 501147.21180226485,
     "upload_bytes_per_second": 2178019.4106091666,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.72847052603537,
+    "estimated_full_shards_per_hour": 107.5345255427452,
     "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

@@ -136,3 +136,4 @@
 {"step": 658600, "loss": 0.5846948027610779, "lr": 0.0001, "grad_norm": 0.1931559684828068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96401125533, "tokens_per_second": 150968.4826024349, "data_wait_s": 1.0240999472443946e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 658700, "loss": 0.5515803694725037, "lr": 0.0001, "grad_norm": 0.1961138554767908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96415753876, "tokens_per_second": 153168.6945994227, "data_wait_s": 8.998998964671046e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 658800, "loss": 0.5704889893531799, "lr": 0.0001, "grad_norm": 0.19855481073692224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96430477585, "tokens_per_second": 153797.17199487513, "data_wait_s": 1.0458999895490706e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 658900, "loss": 0.5716390013694763, "lr": 0.0001, "grad_norm": 0.19920145486725915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96445199974, "tokens_per_second": 153532.37566865486, "data_wait_s": 1.3941998986410908e-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 sha256:673fa60f374a9ab6a1f7f84903e5550cda6dab35d5bd914322631ad63d61c78f
-size 294015
+oid sha256:5d81e1ce413ae35553e694ed5fb2273f0b89c16894f89fcc1c3542cc0a50e517
+size 296133