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": 76774.66448643194,
-      "required_pages_per_second_upper_estimate": 0.005311449430936116
+      "required_token_positions_per_second": 76729.7081701617,
+      "required_pages_per_second_upper_estimate": 0.005308339248663515
     },
     {
       "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": 9596.833060803992,
-      "required_pages_per_second_upper_estimate": 0.004017993597895548
+      "required_token_positions_per_second": 9591.213521270212,
+      "required_pages_per_second_upper_estimate": 0.004015640814042084
     },
     {
       "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": 19193.666121607985,
-      "required_pages_per_second_upper_estimate": 0.00399753448211679
+      "required_token_positions_per_second": 19182.427042540425,
+      "required_pages_per_second_upper_estimate": 0.003995193678341464
     },
     {
       "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": 28790.49918241198,
-      "required_pages_per_second_upper_estimate": 0.0017259464654387447
+      "required_token_positions_per_second": 28773.64056381064,
+      "required_pages_per_second_upper_estimate": 0.0017249358169952147
     },
     {
       "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": 28790.49918241198,
-      "required_pages_per_second_upper_estimate": 0.0022383346754099353
+      "required_token_positions_per_second": 28773.64056381064,
+      "required_pages_per_second_upper_estimate": 0.002237023991966908
     },
     {
       "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": 28790.49918241198,
-      "required_pages_per_second_upper_estimate": 0.36929361068241784
+      "required_token_positions_per_second": 28773.64056381064,
+      "required_pages_per_second_upper_estimate": 0.3690773664243742
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.42592822528894203,
-    "producer_p95_seconds": 0.3165926900001068,
-    "gpu_update_seconds": 0.9603167984280756,
-    "producer_batches_per_second": 2.3478134122753147,
-    "gpu_batches_per_second": 1.0413230317712665,
-    "producer_headroom": 2.2546446593827256,
+    "producer_batch_seconds": 0.08675086142201849,
+    "producer_p95_seconds": 0.31190920299923164,
+    "gpu_update_seconds": 0.9608794527993657,
+    "producer_batches_per_second": 11.52726305662006,
+    "gpu_batches_per_second": 1.0407132727072714,
+    "producer_headroom": 11.076310218119426,
     "sustainable": true,
-    "buffer_seconds": 60.499958300968764,
+    "buffer_seconds": 60.53540552636004,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296207.6517557955,
-    "upload_bytes_per_second": 3175563.9322219808,
+    "generation_bytes_per_second": 294463.37311084993,
+    "upload_bytes_per_second": 3241646.9512518463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.55926670556449,
+    "estimated_full_shards_per_hour": 63.184985113087876,
     "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": 1.3961634170009347,
-      "bytes": 973137,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 2.0744263130000036,
       "bytes": 17948904,
@@ -562,6 +556,12 @@
       "bytes": 2663289,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.5531271910003852,
+      "bytes": 10487346,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -305,3 +305,4 @@
 {"step": 830500, "loss": 0.5266762375831604, "lr": 3e-05, "grad_norm": 0.29181261907840583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121561936684, "tokens_per_second": 150365.47665462186, "data_wait_s": 1.2590000551426783e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 830600, "loss": 0.527016818523407, "lr": 3e-05, "grad_norm": 0.2842183373083332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121576469306, "tokens_per_second": 150481.12139268251, "data_wait_s": 1.2268003047211096e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 830700, "loss": 0.542384684085846, "lr": 3e-05, "grad_norm": 0.2861808411074264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121591172028, "tokens_per_second": 152185.72819809627, "data_wait_s": 1.1500997061375529e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 830800, "loss": 0.5325660109519958, "lr": 3e-05, "grad_norm": 0.29727016415077706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121605804702, "tokens_per_second": 151587.4508602347, "data_wait_s": 1.5562000044155866e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:394bacab2a9b0d40e2e53ccb3ec60c9c1944f2416dc1b22ce03e7a11e13626dc
-size 655089
+oid sha256:11e92e441e7dc0d991326ef1a6ef00561d6cbfa413e758ba8b80dcfbdae4179b
+size 657207