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": 77052.05607909137,
-      "required_pages_per_second_upper_estimate": 0.005330640024953466
+      "required_token_positions_per_second": 76924.04435752601,
+      "required_pages_per_second_upper_estimate": 0.005321783877027458
     },
     {
       "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": 9631.507009886422,
-      "required_pages_per_second_upper_estimate": 0.004032510856301967
+      "required_token_positions_per_second": 9615.505544690752,
+      "required_pages_per_second_upper_estimate": 0.0040258113769731305
     },
     {
       "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": 19263.014019772843,
-      "required_pages_per_second_upper_estimate": 0.004011977820477472
+      "required_token_positions_per_second": 19231.011089381504,
+      "required_pages_per_second_upper_estimate": 0.004005312454050986
     },
     {
       "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": 28894.521029659267,
-      "required_pages_per_second_upper_estimate": 0.0017321824163490556
+      "required_token_positions_per_second": 28846.516634072254,
+      "required_pages_per_second_upper_estimate": 0.0017293046261320822
     },
     {
       "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": 28894.521029659267,
-      "required_pages_per_second_upper_estimate": 0.002246421916489661
+      "required_token_positions_per_second": 28846.516634072254,
+      "required_pages_per_second_upper_estimate": 0.0022426897858817915
     },
     {
       "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": 28894.521029659267,
-      "required_pages_per_second_upper_estimate": 0.3706278912489484
+      "required_token_positions_per_second": 28846.516634072254,
+      "required_pages_per_second_upper_estimate": 0.3700121424054624
     }
   ],
   "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": 0.3612854624609838,
-    "producer_p95_seconds": 0.1833698740001637,
-    "gpu_update_seconds": 0.9568596057231835,
-    "producer_batches_per_second": 2.7678943768959225,
-    "gpu_batches_per_second": 1.0450853960380233,
-    "producer_headroom": 2.648486322160049,
+    "producer_batch_seconds": 0.06777656743753369,
+    "producer_p95_seconds": 0.17132815900004061,
+    "gpu_update_seconds": 0.9584519458874068,
+    "producer_batches_per_second": 14.754361836362554,
+    "gpu_batches_per_second": 1.0433491259430068,
+    "producer_headroom": 14.141346812388582,
     "sustainable": true,
-    "buffer_seconds": 60.28215516056056,
+    "buffer_seconds": 60.38247259090663,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340789.8806861911,
-    "upload_bytes_per_second": 4127433.343550122,
+    "generation_bytes_per_second": 327212.7967588068,
+    "upload_bytes_per_second": 4197608.118588127,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.12557521285342,
+    "estimated_full_shards_per_hour": 70.21224906037476,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 2301339,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.4037659219993657,
+      "bytes": 10492690,
+      "pages": 42,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -66,3 +66,4 @@
 {"step": 531600, "loss": 0.5742305517196655, "lr": 0.0003, "grad_norm": 0.17469050618035542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77808923030, "tokens_per_second": 153622.97289850537, "data_wait_s": 1.7005999325192533e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 531700, "loss": 0.601482093334198, "lr": 0.0003, "grad_norm": 0.16568858733734793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77823771755, "tokens_per_second": 155227.28266635895, "data_wait_s": 6.833999577793293e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 531800, "loss": 0.6045354008674622, "lr": 0.0003, "grad_norm": 0.18553475762520746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77838537453, "tokens_per_second": 154608.2630370454, "data_wait_s": 1.0071999895444605e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 531900, "loss": 0.639557957649231, "lr": 0.0003, "grad_norm": 0.180102644276118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77853197157, "tokens_per_second": 153762.32467936815, "data_wait_s": 1.095200059353374e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a978f4e74455ac4b35121cde02aa5aee1c981487347b4a6e1a142de1c389ec95
-size 144972
+oid sha256:2a692d7e08ad3b5eba55c4968a770e46850b8abebe9c474df7b16f567854a6f7
+size 147090