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": 75075.29412071979,
-      "required_pages_per_second_upper_estimate": 0.006507336217790635
+      "required_token_positions_per_second": 75960.78953772073,
+      "required_pages_per_second_upper_estimate": 0.006584088716268661
     },
     {
       "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": 9384.411765089973,
-      "required_pages_per_second_upper_estimate": 0.003743582162553843
+      "required_token_positions_per_second": 9495.098692215091,
+      "required_pages_per_second_upper_estimate": 0.00378773683270109
     },
     {
       "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": 18768.823530179947,
-      "required_pages_per_second_upper_estimate": 0.004341837003694829
+      "required_token_positions_per_second": 18990.197384430183,
+      "required_pages_per_second_upper_estimate": 0.004393047948828853
     },
     {
       "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": 28153.23529526992,
-      "required_pages_per_second_upper_estimate": 0.0017143843208718666
+      "required_token_positions_per_second": 28485.296076645274,
+      "required_pages_per_second_upper_estimate": 0.0017346050802693493
     },
     {
       "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": 28153.23529526992,
-      "required_pages_per_second_upper_estimate": 0.0021673139861731186
+      "required_token_positions_per_second": 28485.296076645274,
+      "required_pages_per_second_upper_estimate": 0.002192876944326468
     },
     {
       "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": 28153.23529526992,
-      "required_pages_per_second_upper_estimate": 0.3495689595498953
+      "required_token_positions_per_second": 28485.296076645274,
+      "required_pages_per_second_upper_estimate": 0.3536920431186321
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.47472585217201413,
-    "producer_p95_seconds": 0.2778958380004042,
-    "gpu_update_seconds": 0.9820540946725649,
-    "producer_batches_per_second": 2.1064789192008355,
-    "gpu_batches_per_second": 1.0182738460384084,
-    "producer_headroom": 2.0686762479426197,
+    "producer_batch_seconds": 0.11523102538285457,
+    "producer_p95_seconds": 0.3124006370017014,
+    "gpu_update_seconds": 0.970606025143907,
+    "producer_batches_per_second": 8.678218358966298,
+    "gpu_batches_per_second": 1.0302841462907,
+    "producer_headroom": 8.423131026727159,
     "sustainable": true,
-    "buffer_seconds": 60.88735386969902,
+    "buffer_seconds": 60.177573558922234,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266903.21377771767,
-    "upload_bytes_per_second": 5147434.733777678,
+    "generation_bytes_per_second": 283906.5385168827,
+    "upload_bytes_per_second": 5213977.019993049,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.27121648787163,
+    "estimated_full_shards_per_hour": 60.91973415975438,
     "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": 0.929622351000944,
-      "bytes": 6601933,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 2.1250439309987996,
       "bytes": 6028843,
@@ -562,6 +556,12 @@
       "bytes": 582089,
       "pages": 18,
       "shards": 1
+    },
+    {
+      "seconds": 1.208456311996997,
+      "bytes": 13456782,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -232,3 +232,4 @@
 {"step": 378200, "loss": 0.5748591497540474, "lr": 0.0003, "grad_norm": 0.15164900917258697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55357767242, "tokens_per_second": 149328.52792212137, "data_wait_s": 1.9019000319531187e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 378300, "loss": 0.557722071185708, "lr": 0.0003, "grad_norm": 0.163479854615462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55372456151, "tokens_per_second": 150226.34467800736, "data_wait_s": 1.0760002624010667e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 378400, "loss": 0.5520606003701687, "lr": 0.0003, "grad_norm": 0.1735362186521396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55386921035, "tokens_per_second": 122804.24020786358, "data_wait_s": 1.145099668065086e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 378500, "loss": 0.6116420533508062, "lr": 0.0003, "grad_norm": 0.17059345758723035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55401541945, "tokens_per_second": 148714.45944499297, "data_wait_s": 8.55699909152463e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd766e98612d006291d29cf8374ac0c088038c187b6f6004bd38e769e27e3865
-size 498909
+oid sha256:b465a4c1343bf9422c66aaee8f2d1558d6c748e5c070f95e424ba1916bdd01b1
+size 501027