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": 76791.74846629043,
-      "required_pages_per_second_upper_estimate": 0.005312631340303021
+      "required_token_positions_per_second": 76473.163323782,
+      "required_pages_per_second_upper_estimate": 0.005290590880924912
     },
     {
       "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": 9598.968558286304,
-      "required_pages_per_second_upper_estimate": 0.004018887686097134
+      "required_token_positions_per_second": 9559.14541547275,
+      "required_pages_per_second_upper_estimate": 0.004002214567802885
     },
     {
       "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": 19197.937116572608,
-      "required_pages_per_second_upper_estimate": 0.0039984240177342095
+      "required_token_positions_per_second": 19118.2908309455,
+      "required_pages_per_second_upper_estimate": 0.003981835796851882
     },
     {
       "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": 28796.905674858914,
-      "required_pages_per_second_upper_estimate": 0.001726330524878791
+      "required_token_positions_per_second": 28677.436246418252,
+      "required_pages_per_second_upper_estimate": 0.0017191685150630281
     },
     {
       "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": 28796.905674858914,
-      "required_pages_per_second_upper_estimate": 0.0022388327520185022
+      "required_token_positions_per_second": 28677.436246418252,
+      "required_pages_per_second_upper_estimate": 0.002229544529447721
     },
     {
       "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": 28796.905674858914,
-      "required_pages_per_second_upper_estimate": 0.3693757862887715
+      "required_token_positions_per_second": 28677.436246418252,
+      "required_pages_per_second_upper_estimate": 0.3678433607370128
     }
   ],
   "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": 0.3467788346640077,
-    "producer_p95_seconds": 0.1573574469994128,
-    "gpu_update_seconds": 0.9601031552545605,
-    "producer_batches_per_second": 2.88368233594445,
-    "gpu_batches_per_second": 1.041554748077941,
-    "producer_headroom": 2.768632509492108,
+    "producer_batch_seconds": 0.06387275091406508,
+    "producer_p95_seconds": 0.16257011400011834,
+    "gpu_update_seconds": 0.964102919188014,
+    "producer_batches_per_second": 15.656128563264923,
+    "gpu_batches_per_second": 1.0372336605330674,
+    "producer_headroom": 15.09411925102656,
     "sustainable": true,
-    "buffer_seconds": 60.48649878103731,
+    "buffer_seconds": 60.738483908844884,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399886.51730805356,
-    "upload_bytes_per_second": 5676127.341662556,
+    "generation_bytes_per_second": 393551.4104153936,
+    "upload_bytes_per_second": 5685116.03939964,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.80633773261266,
+    "estimated_full_shards_per_hour": 84.4469712671886,
     "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": [
@@ -412,6 +412,12 @@
       "bytes": 10744820,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0739986469998257,
+      "bytes": 6490657,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 561500, "loss": 0.5722002387046814, "lr": 0.0003, "grad_norm": 0.1494081318000216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82186091670, "tokens_per_second": 151706.1245322584, "data_wait_s": 1.5481999980693217e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 561600, "loss": 0.592818021774292, "lr": 0.0003, "grad_norm": 0.15642986707509363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82200627359, "tokens_per_second": 150529.44197882366, "data_wait_s": 1.7706999642541632e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 561700, "loss": 0.582870602607727, "lr": 0.0003, "grad_norm": 0.1587683086837042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82215218311, "tokens_per_second": 151894.4562989931, "data_wait_s": 1.654700008657528e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 561800, "loss": 0.567470908164978, "lr": 0.0003, "grad_norm": 0.17873162162093043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82229882507, "tokens_per_second": 152559.57835867302, "data_wait_s": 1.4860999726806767e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13aebd0a72839ab4f4f737e1f51315934371c40152bc075a596932888a3b9061
-size 98376
+oid sha256:bd00c15b5c614c189611abe50b8ec62c76872b3ad42e97466296c2b4755b2713
+size 100494