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": 76930.90350154375,
-      "required_pages_per_second_upper_estimate": 0.005322258408526012
+      "required_token_positions_per_second": 76856.3238130518,
+      "required_pages_per_second_upper_estimate": 0.00531709881262742
     },
     {
       "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": 9616.36293769297,
-      "required_pages_per_second_upper_estimate": 0.004026170349518758
+      "required_token_positions_per_second": 9607.040476631475,
+      "required_pages_per_second_upper_estimate": 0.004022267229747434
     },
     {
       "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": 19232.72587538594,
-      "required_pages_per_second_upper_estimate": 0.004005669598753761
+      "required_token_positions_per_second": 19214.08095326295,
+      "required_pages_per_second_upper_estimate": 0.004001786353175205
     },
     {
       "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": 28849.088813078906,
-      "required_pages_per_second_upper_estimate": 0.001729458824336041
+      "required_token_positions_per_second": 28821.12142989443,
+      "required_pages_per_second_upper_estimate": 0.0017277822224178532
     },
     {
       "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": 28849.088813078906,
-      "required_pages_per_second_upper_estimate": 0.002242889761485739
+      "required_token_positions_per_second": 28821.12142989443,
+      "required_pages_per_second_upper_estimate": 0.002240715420458664
     },
     {
       "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": 28849.088813078906,
-      "required_pages_per_second_upper_estimate": 0.3700451355559691
+      "required_token_positions_per_second": 28821.12142989443,
+      "required_pages_per_second_upper_estimate": 0.36968639999351505
     }
   ],
   "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": 0.08762669257029643,
-    "producer_p95_seconds": 0.17802224599836336,
-    "gpu_update_seconds": 0.9583664905030073,
-    "producer_batches_per_second": 11.412047752433129,
-    "gpu_batches_per_second": 1.0434421590378655,
-    "producer_headroom": 10.936924153952068,
+    "producer_batch_seconds": 0.07195431516400674,
+    "producer_p95_seconds": 0.20641076499850897,
+    "gpu_update_seconds": 0.9592964682950325,
+    "producer_batches_per_second": 13.89770714543919,
+    "gpu_batches_per_second": 1.0424306072733807,
+    "producer_headroom": 13.332021382018452,
     "sustainable": true,
-    "buffer_seconds": 60.377088901689454,
+    "buffer_seconds": 60.43567750258705,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320170.5308558638,
-    "upload_bytes_per_second": 5423318.807282602,
+    "generation_bytes_per_second": 320072.301174809,
+    "upload_bytes_per_second": 5429452.783823526,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.70114273316322,
+    "estimated_full_shards_per_hour": 68.6800649302788,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 6639528,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.0738694510000641,
+      "bytes": 6116466,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -81,3 +81,4 @@
 {"step": 768100, "loss": 0.5042897462844849, "lr": 3e-05, "grad_norm": 0.2413063474768963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112427227492, "tokens_per_second": 150296.4566534407, "data_wait_s": 1.1320000339765102e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 768200, "loss": 0.5287339091300964, "lr": 3e-05, "grad_norm": 0.2479985593271508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112441856491, "tokens_per_second": 151739.89184902445, "data_wait_s": 1.1863001418532804e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 768300, "loss": 0.5232650637626648, "lr": 3e-05, "grad_norm": 0.24733559985406833, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112456519879, "tokens_per_second": 152929.25510498046, "data_wait_s": 1.6621001122985035e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 768400, "loss": 0.5567591190338135, "lr": 3e-05, "grad_norm": 0.25034149847708914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112471228766, "tokens_per_second": 152735.4847341141, "data_wait_s": 1.2127000445616432e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ddda5a923b760dbe37193d5d4528e080559d795233f89eb35f2f21e2fcab9f9
-size 176742
+oid sha256:41b5052a9c71a91c637078ac3bd1e679af1aba1de7f210f9cee8e3974f3f1add
+size 178860