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": 77383.47630712598,
-      "required_pages_per_second_upper_estimate": 0.005353568445329777
+      "required_token_positions_per_second": 77305.24273554547,
+      "required_pages_per_second_upper_estimate": 0.005348156065320942
     },
     {
       "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": 9672.934538390748,
-      "required_pages_per_second_upper_estimate": 0.0040498556973815585
+      "required_token_positions_per_second": 9663.155341943184,
+      "required_pages_per_second_upper_estimate": 0.004045761352041807
     },
     {
       "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": 19345.869076781495,
-      "required_pages_per_second_upper_estimate": 0.004029234343817583
+      "required_token_positions_per_second": 19326.310683886368,
+      "required_pages_per_second_upper_estimate": 0.0040251608463670345
     },
     {
       "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": 29018.803615172244,
-      "required_pages_per_second_upper_estimate": 0.0017396329675820387
+      "required_token_positions_per_second": 28989.46602582955,
+      "required_pages_per_second_upper_estimate": 0.0017378742238968493
     },
     {
       "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": 29018.803615172244,
-      "required_pages_per_second_upper_estimate": 0.002256084340851974
+      "required_token_positions_per_second": 28989.46602582955,
+      "required_pages_per_second_upper_estimate": 0.0022538034723229975
     },
     {
       "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": 29018.803615172244,
-      "required_pages_per_second_upper_estimate": 0.37222205481166537
+      "required_token_positions_per_second": 28989.46602582955,
+      "required_pages_per_second_upper_estimate": 0.3718457437158265
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07381040456237997,
-    "producer_p95_seconds": 0.1910857720013155,
-    "gpu_update_seconds": 0.9527615392643021,
-    "producer_batches_per_second": 13.5482254287722,
-    "gpu_batches_per_second": 1.0495805705719128,
-    "producer_headroom": 12.908228113816762,
+    "producer_batch_seconds": 0.09629861364838632,
+    "producer_p95_seconds": 0.21305067499997676,
+    "gpu_update_seconds": 0.9537257421494308,
+    "producer_batches_per_second": 10.384365486831253,
+    "gpu_batches_per_second": 1.0485194598462655,
+    "producer_headroom": 9.903836680679072,
     "sustainable": true,
-    "buffer_seconds": 60.02397697365103,
+    "buffer_seconds": 60.084721755414144,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361788.24125972635,
-    "upload_bytes_per_second": 3793603.2912303167,
+    "generation_bytes_per_second": 385193.74299153517,
+    "upload_bytes_per_second": 3833374.637762791,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.63133457511752,
+    "estimated_full_shards_per_hour": 82.65361039456884,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 6108290,
       "pages": 9,
       "shards": 2
+    },
+    {
+      "seconds": 1.1298799779997353,
+      "bytes": 7635963,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -100,3 +100,4 @@
 {"step": 695000, "loss": 0.546134889125824, "lr": 8e-05, "grad_norm": 0.21493580285015115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101729356649, "tokens_per_second": 152240.4627242544, "data_wait_s": 1.3352000678423792e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 695100, "loss": 0.5333627462387085, "lr": 8e-05, "grad_norm": 0.2076473975651665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101743898324, "tokens_per_second": 142495.10113668753, "data_wait_s": 1.2911999874631874e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 695200, "loss": 0.5591259002685547, "lr": 8e-05, "grad_norm": 0.19911689503341654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101758520827, "tokens_per_second": 152081.54863085336, "data_wait_s": 1.541399979032576e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 695300, "loss": 0.5476700067520142, "lr": 8e-05, "grad_norm": 0.20254950488096524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101773175001, "tokens_per_second": 152781.06473411171, "data_wait_s": 1.7416001355741173e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59702e00593800e4895150459290b879fbd831df352dbea2330fe9379892f3cb
-size 217767
+oid sha256:d798c2ff8594a1a06412d7e73dd831a88959eab0349113e80fa8062c3621e2a7
+size 219885