Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75491.12986909541,
-      "required_pages_per_second_upper_estimate": 0.006543379806533766
+      "required_token_positions_per_second": 75021.71285716708,
+      "required_pages_per_second_upper_estimate": 0.0065026919296663575
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9436.391233636927,
-      "required_pages_per_second_upper_estimate": 0.0037643175497195334
+      "required_token_positions_per_second": 9377.714107145885,
+      "required_pages_per_second_upper_estimate": 0.003740910366661036
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18872.782467273853,
-      "required_pages_per_second_upper_estimate": 0.004365886074176701
+      "required_token_positions_per_second": 18755.42821429177,
+      "required_pages_per_second_upper_estimate": 0.0043387382331136065
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28309.17370091078,
-      "required_pages_per_second_upper_estimate": 0.0017238801516298102
+      "required_token_positions_per_second": 28133.142321437652,
+      "required_pages_per_second_upper_estimate": 0.001713160764185168
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28309.17370091078,
-      "required_pages_per_second_upper_estimate": 0.002179318556304487
+      "required_token_positions_per_second": 28133.142321437652,
+      "required_pages_per_second_upper_estimate": 0.0021657671734265993
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28309.17370091078,
-      "required_pages_per_second_upper_estimate": 0.3515051926556835
+      "required_token_positions_per_second": 28133.142321437652,
+      "required_pages_per_second_upper_estimate": 0.3493194720617561
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07747386432816938,
-    "producer_p95_seconds": 0.22992031600006158,
-    "gpu_update_seconds": 0.9766445425819862,
-    "producer_batches_per_second": 12.907578686976654,
-    "gpu_batches_per_second": 1.0239139793442844,
-    "producer_headroom": 12.606116282583308,
+    "producer_batch_seconds": 0.19619156360161583,
+    "producer_p95_seconds": 0.19941095000103815,
+    "gpu_update_seconds": 0.9827554876062327,
+    "producer_batches_per_second": 5.097059127530008,
+    "gpu_batches_per_second": 1.0175471036399615,
+    "producer_headroom": 5.009162828233552,
     "sustainable": true,
-    "buffer_seconds": 60.55196164008314,
+    "buffer_seconds": 60.93084023158642,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356147.2979062609,
-    "upload_bytes_per_second": 3612953.2759696795,
+    "generation_bytes_per_second": 333271.7368757514,
+    "upload_bytes_per_second": 4967026.690318246,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.42091944590445,
+    "estimated_full_shards_per_hour": 71.51235656456382,
     "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": 33.908926936999706,
-      "bytes": 9910795,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 2.0074904029997924,
       "bytes": 13527574,
@@ -562,6 +556,12 @@
       "bytes": 8238622,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.0565556260007725,
+      "bytes": 5902213,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -120,3 +120,4 @@
 {"step": 249100, "loss": 0.5791001301258802, "lr": 0.0003, "grad_norm": 0.15355182679166174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36461575458, "tokens_per_second": 149025.73628882563, "data_wait_s": 1.1598000128287822e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 249200, "loss": 0.6431687735021114, "lr": 0.0003, "grad_norm": 0.13891328064437083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36476268973, "tokens_per_second": 121536.00045341575, "data_wait_s": 1.3046999811194837e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 249300, "loss": 0.5752866845577955, "lr": 0.0003, "grad_norm": 0.16293882183368283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36491026020, "tokens_per_second": 150820.65796480002, "data_wait_s": 1.0529000064707361e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 249400, "loss": 0.6120243947952986, "lr": 0.0003, "grad_norm": 0.14405296487039118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36505670984, "tokens_per_second": 149597.77159190035, "data_wait_s": 1.2959999367012642e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1903d53bdf332ed35c43bfb7d7a59f41ed92396580987dc807201d49f3d7578b
-size 260127
+oid sha256:faea2d217aa3063efb67bf5cb5e3341c19ff3ad4b1235a2854a1458741031c36
+size 262245