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": 77063.56796846028,
-      "required_pages_per_second_upper_estimate": 0.00533143644417127
+      "required_token_positions_per_second": 77466.69650869047,
+      "required_pages_per_second_upper_estimate": 0.0053593258119714795
     },
     {
       "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": 9632.945996057535,
-      "required_pages_per_second_upper_estimate": 0.004033113329762364
+      "required_token_positions_per_second": 9683.337063586308,
+      "required_pages_per_second_upper_estimate": 0.004054211017451512
     },
     {
       "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": 19265.89199211507,
-      "required_pages_per_second_upper_estimate": 0.004012577226219123
+      "required_token_positions_per_second": 19366.674127172617,
+      "required_pages_per_second_upper_estimate": 0.004033567487147979
     },
     {
       "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": 28898.837988172603,
-      "required_pages_per_second_upper_estimate": 0.0017324412114203163
+      "required_token_positions_per_second": 29050.011190758927,
+      "required_pages_per_second_upper_estimate": 0.0017415038140872513
     },
     {
       "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": 28898.837988172603,
-      "required_pages_per_second_upper_estimate": 0.0022467575410257787
+      "required_token_positions_per_second": 29050.011190758927,
+      "required_pages_per_second_upper_estimate": 0.0022585105925862206
     },
     {
       "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": 28898.837988172603,
-      "required_pages_per_second_upper_estimate": 0.37068326455756856
+      "required_token_positions_per_second": 29050.011190758927,
+      "required_pages_per_second_upper_estimate": 0.3726223520832073
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10966553089045306,
-    "producer_p95_seconds": 0.2304532050002308,
-    "gpu_update_seconds": 0.9567166683766132,
-    "producer_batches_per_second": 9.118635471695464,
-    "gpu_batches_per_second": 1.0452415360305485,
-    "producer_headroom": 8.723950548621293,
+    "producer_batch_seconds": 0.37647408068730215,
+    "producer_p95_seconds": 0.2351292699968326,
+    "gpu_update_seconds": 0.9517380154674461,
+    "producer_batches_per_second": 2.6562253586604703,
+    "gpu_batches_per_second": 1.0507093167953894,
+    "producer_headroom": 2.528030651485821,
     "sustainable": true,
-    "buffer_seconds": 60.27315010772663,
+    "buffer_seconds": 60.91123298991655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 410182.87243185553,
+    "generation_bytes_per_second": 384662.4960782209,
     "upload_bytes_per_second": 1870651.0777451813,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.01569585530042,
+    "estimated_full_shards_per_hour": 82.5396171737668,
     "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

@@ -176,3 +176,4 @@
 {"step": 662600, "loss": 0.48360851407051086, "lr": 0.0001, "grad_norm": 0.18846363952215442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96987475711, "tokens_per_second": 153696.57958388873, "data_wait_s": 1.367199729429558e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 662700, "loss": 0.5400761365890503, "lr": 0.0001, "grad_norm": 0.20251129177201113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97002009539, "tokens_per_second": 151179.20269854466, "data_wait_s": 8.656999852973968e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 662800, "loss": 0.5265199542045593, "lr": 0.0001, "grad_norm": 0.20550581491649966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97016461604, "tokens_per_second": 150936.0243713266, "data_wait_s": 1.0705000022426248e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 662900, "loss": 0.5354124307632446, "lr": 0.0001, "grad_norm": 0.1932775049472148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97031185228, "tokens_per_second": 153468.054697483, "data_wait_s": 1.001499913400039e-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 sha256:60de15697c45b5254e50ca1b097d6a7716bd8006792be3ddc35828d82f363bbf
-size 379518
+oid sha256:69bb04f76dbde324a7ea0db5f61ffc13d7a80c40250e3fba265cb75101cc38a7
+size 381636