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": 75641.7297664733,
-      "required_pages_per_second_upper_estimate": 0.005233070378238653
+      "required_token_positions_per_second": 77301.82519888892,
+      "required_pages_per_second_upper_estimate": 0.005347919632205264
     },
     {
       "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": 9455.216220809163,
-      "required_pages_per_second_upper_estimate": 0.0039587015842856175
+      "required_token_positions_per_second": 9662.728149861116,
+      "required_pages_per_second_upper_estimate": 0.004045582495637831
     },
     {
       "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": 18910.432441618326,
-      "required_pages_per_second_upper_estimate": 0.003938544375949379
+      "required_token_positions_per_second": 19325.45629972223,
+      "required_pages_per_second_upper_estimate": 0.0040249829006772705
     },
     {
       "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": 28365.648662427488,
-      "required_pages_per_second_upper_estimate": 0.001700477325474857
+      "required_token_positions_per_second": 28988.184449583343,
+      "required_pages_per_second_upper_estimate": 0.0017377973953577428
     },
     {
       "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": 28365.648662427488,
-      "required_pages_per_second_upper_estimate": 0.002205304416201086
+      "required_token_positions_per_second": 28988.184449583343,
+      "required_pages_per_second_upper_estimate": 0.002253703835406913
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28365.648662427488,
-      "required_pages_per_second_upper_estimate": 0.36384408438100446
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28988.184449583343,
+      "required_pages_per_second_upper_estimate": 0.3718293050317895
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09976324890646993,
-    "producer_p95_seconds": 0.35383229500439484,
-    "gpu_update_seconds": 0.974700079276591,
-    "producer_batches_per_second": 10.023731293449758,
-    "gpu_batches_per_second": 1.0259566211815498,
-    "producer_headroom": 9.770131686372725,
+    "producer_batch_seconds": 0.12107199702376192,
+    "producer_p95_seconds": 0.35173528399900533,
+    "gpu_update_seconds": 0.9537679066478202,
+    "producer_batches_per_second": 8.259548240570751,
+    "gpu_batches_per_second": 1.0484731065387494,
+    "producer_headroom": 7.877692035265851,
     "sustainable": true,
-    "buffer_seconds": 60.43140491514864,
+    "buffer_seconds": 60.08737811881267,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298861.00976301555,
-    "upload_bytes_per_second": 4654394.899486298,
+    "generation_bytes_per_second": 325601.3383442341,
+    "upload_bytes_per_second": 4645292.19340822,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.12861556690073,
+    "estimated_full_shards_per_hour": 69.86646759743945,
     "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": [
-    {
-      "seconds": 1.6370693080025376,
-      "bytes": 5715454,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.2990395020024152,
       "bytes": 2748829,
@@ -562,6 +556,12 @@
       "bytes": 7339502,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.8050662450041273,
+      "bytes": 1009695,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -430,3 +430,4 @@
 {"step": 688000, "loss": 0.4782137870788574, "lr": 0.0001, "grad_norm": 0.20103014709600134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100704686333, "tokens_per_second": 151602.26420412486, "data_wait_s": 1.779800368240103e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688100, "loss": 0.5473445057868958, "lr": 0.0001, "grad_norm": 0.20208139775954997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100719252292, "tokens_per_second": 151831.41347636824, "data_wait_s": 9.546994988340884e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 688200, "loss": 0.5151166915893555, "lr": 0.0001, "grad_norm": 0.2120875927685729, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100733890275, "tokens_per_second": 153225.4642736849, "data_wait_s": 1.7035999917425215e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 688300, "loss": 0.5191828608512878, "lr": 0.0001, "grad_norm": 0.1867570862786017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100748617566, "tokens_per_second": 153053.5687605134, "data_wait_s": 1.6682999557815492e-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 sha256:5037f035d3e12b4e87ddcd141b8cd548fef149b7f6c442101c639c1c4dba7aa4
-size 921405
+oid sha256:30e4d8faf4fcf46f12cdbff53a078e461c8aa9cf90f062eb269a391a1b946817
+size 923523