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": 77026.31621935265,
-      "required_pages_per_second_upper_estimate": 0.005328859281732038
+      "required_token_positions_per_second": 77186.4128462336,
+      "required_pages_per_second_upper_estimate": 0.005339935137855015
     },
     {
       "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": 9628.289527419081,
-      "required_pages_per_second_upper_estimate": 0.004031163763581566
+      "required_token_positions_per_second": 9648.3016057792,
+      "required_pages_per_second_upper_estimate": 0.00403954240289123
     },
     {
       "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": 19256.579054838163,
-      "required_pages_per_second_upper_estimate": 0.004010637586983015
+      "required_token_positions_per_second": 19296.6032115584,
+      "required_pages_per_second_upper_estimate": 0.004018973563319848
     },
     {
       "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": 28884.86858225724,
-      "required_pages_per_second_upper_estimate": 0.001731603766865738
+      "required_token_positions_per_second": 28944.9048173376,
+      "required_pages_per_second_upper_estimate": 0.0017352028474887824
     },
     {
       "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": 28884.86858225724,
-      "required_pages_per_second_upper_estimate": 0.0022456714811607825
+      "required_token_positions_per_second": 28944.9048173376,
+      "required_pages_per_second_upper_estimate": 0.00225033903436679
     },
     {
       "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": 28884.86858225724,
-      "required_pages_per_second_upper_estimate": 0.37050408001766577
+      "required_token_positions_per_second": 28944.9048173376,
+      "required_pages_per_second_upper_estimate": 0.3712741603793897
     }
   ],
   "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": 0.4104763055547167,
-    "producer_p95_seconds": 0.17498687300030724,
-    "gpu_update_seconds": 0.9571793591951117,
-    "producer_batches_per_second": 2.4361942125955416,
-    "gpu_batches_per_second": 1.0447362768466883,
-    "producer_headroom": 2.33187481528704,
+    "producer_batch_seconds": 0.10275180482815927,
+    "producer_p95_seconds": 0.2089943620003396,
+    "gpu_update_seconds": 0.9551940203112788,
+    "producer_batches_per_second": 9.732189149109221,
+    "gpu_batches_per_second": 1.0469077263215276,
+    "producer_headroom": 9.296128879767439,
     "sustainable": true,
-    "buffer_seconds": 60.30229962929204,
+    "buffer_seconds": 60.17722327961056,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336240.6925314308,
-    "upload_bytes_per_second": 4303635.813146845,
+    "generation_bytes_per_second": 336071.2119055927,
+    "upload_bytes_per_second": 4325951.494597062,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.14942533452218,
+    "estimated_full_shards_per_hour": 72.11305873752437,
     "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": [
@@ -508,6 +508,12 @@
       "bytes": 5935208,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1118033149996336,
+      "bytes": 6738611,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -70,3 +70,4 @@
 {"step": 447200, "loss": 0.6125456094741821, "lr": 0.0003, "grad_norm": 0.18039655206486915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65456727001, "tokens_per_second": 153944.95189737467, "data_wait_s": 1.4642999303760007e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 447300, "loss": 0.563656747341156, "lr": 0.0003, "grad_norm": 0.1657343197335768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65471367768, "tokens_per_second": 152871.88170210554, "data_wait_s": 2.185199991799891e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 447400, "loss": 0.5943614840507507, "lr": 0.0003, "grad_norm": 0.1546951696609428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65486051783, "tokens_per_second": 152379.3781723956, "data_wait_s": 1.3396999747783411e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 447500, "loss": 0.6167363524436951, "lr": 0.0003, "grad_norm": 0.18652145726225824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65500579390, "tokens_per_second": 151793.33238177557, "data_wait_s": 8.721999620320275e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b1b38cad0fcb2aece4eed499a1dcf30a51f4748f78fa8490127fd3a9b71c9be
-size 153444
+oid sha256:7d91afd6788a6c305feb2fd9574b741a05f0f8039bc60e4da603dafacc040dc5
+size 155562