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": 76779.22893019585,
-      "required_pages_per_second_upper_estimate": 0.005311765209746672
+      "required_token_positions_per_second": 76567.38825141123,
+      "required_pages_per_second_upper_estimate": 0.0052971095800502565
     },
     {
       "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": 9597.403616274481,
-      "required_pages_per_second_upper_estimate": 0.004018232477556489
+      "required_token_positions_per_second": 9570.923531426404,
+      "required_pages_per_second_upper_estimate": 0.00400714581899765
     },
     {
       "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": 19194.807232548963,
-      "required_pages_per_second_upper_estimate": 0.0039977721454326764
+      "required_token_positions_per_second": 19141.84706285281,
+      "required_pages_per_second_upper_estimate": 0.003986741938738564
     },
     {
       "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": 28792.210848823444,
-      "required_pages_per_second_upper_estimate": 0.0017260490772265488
+      "required_token_positions_per_second": 28712.770594279213,
+      "required_pages_per_second_upper_estimate": 0.0017212867552648674
     },
     {
       "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": 28792.210848823444,
-      "required_pages_per_second_upper_estimate": 0.002238467749944602
+      "required_token_positions_per_second": 28712.770594279213,
+      "required_pages_per_second_upper_estimate": 0.0022322916195745396
     },
     {
       "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": 28792.210848823444,
-      "required_pages_per_second_upper_estimate": 0.36931556610129995
+      "required_token_positions_per_second": 28712.770594279213,
+      "required_pages_per_second_upper_estimate": 0.36829659181230634
     }
   ],
   "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": 0.09989558369532858,
-    "producer_p95_seconds": 0.17934741899989604,
-    "gpu_update_seconds": 0.96025970861247,
-    "producer_batches_per_second": 10.010452544627986,
-    "gpu_batches_per_second": 1.0413849410020055,
-    "producer_headroom": 9.612634243583429,
+    "producer_batch_seconds": 0.05051720648435065,
+    "producer_p95_seconds": 0.17147974799991061,
+    "gpu_update_seconds": 0.9629164802893887,
+    "producer_batches_per_second": 19.795235516631003,
+    "gpu_batches_per_second": 1.038511667906511,
+    "producer_headroom": 19.061158510173822,
     "sustainable": true,
-    "buffer_seconds": 60.49636164258561,
+    "buffer_seconds": 60.66373825823149,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360974.41496928997,
-    "upload_bytes_per_second": 5988586.308724129,
+    "generation_bytes_per_second": 360664.09441280673,
+    "upload_bytes_per_second": 5859858.298880527,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.45670639809632,
+    "estimated_full_shards_per_hour": 77.39011883056786,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 7651468,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9403381189995343,
+      "bytes": 1076403,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 489000, "loss": 0.589903712272644, "lr": 0.0003, "grad_norm": 0.17181293917392576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71573585497, "tokens_per_second": 153517.7667324121, "data_wait_s": 1.5919999896141235e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 489100, "loss": 0.5786427855491638, "lr": 0.0003, "grad_norm": 0.1550712551863727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71588238509, "tokens_per_second": 153157.35521647448, "data_wait_s": 1.296700065722689e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 489200, "loss": 0.5819711685180664, "lr": 0.0003, "grad_norm": 0.17574398380406173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71602755677, "tokens_per_second": 151072.58632100123, "data_wait_s": 1.4813999769103248e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 489300, "loss": 0.6015153527259827, "lr": 0.0003, "grad_norm": 0.16216718847017506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71617227678, "tokens_per_second": 150889.70193210422, "data_wait_s": 7.884999831730966e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b03a941215866543c5ac00580ca51e48e39c41d876fefdf846a5423cc6ea171a
-size 89121
+oid sha256:e03b9179d7dea0a7951929ca42933b96134aeaa5fb27a1fe271817c0e08f3a09
+size 91239