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": 74808.33735482603,
-      "required_pages_per_second_upper_estimate": 0.0064841970819187016
+      "required_token_positions_per_second": 74443.61041126814,
+      "required_pages_per_second_upper_estimate": 0.006452583474843091
     },
     {
       "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": 9351.042169353253,
-      "required_pages_per_second_upper_estimate": 0.0037302705318945483
+      "required_token_positions_per_second": 9305.451301408517,
+      "required_pages_per_second_upper_estimate": 0.003712083653027173
     },
     {
       "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": 18702.084338706507,
-      "required_pages_per_second_upper_estimate": 0.0043263980663171015
+      "required_token_positions_per_second": 18610.902602817034,
+      "required_pages_per_second_upper_estimate": 0.004305304776462815
     },
     {
       "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": 28053.12650805976,
-      "required_pages_per_second_upper_estimate": 0.0017082882209609824
+      "required_token_positions_per_second": 27916.35390422555,
+      "required_pages_per_second_upper_estimate": 0.0016999594869778732
     },
     {
       "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": 28053.12650805976,
-      "required_pages_per_second_upper_estimate": 0.002159607334614821
+      "required_token_positions_per_second": 27916.35390422555,
+      "required_pages_per_second_upper_estimate": 0.0021490782009608643
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28053.12650805976,
-      "required_pages_per_second_upper_estimate": 0.34832594345530327
+      "required_token_positions_per_second": 27916.35390422555,
+      "required_pages_per_second_upper_estimate": 0.34662768546414136
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.27917365273438577,
-    "producer_p95_seconds": 0.20891276099973766,
-    "gpu_update_seconds": 0.9855585969020827,
-    "producer_batches_per_second": 3.5819999136932528,
-    "gpu_batches_per_second": 1.0146530131676708,
-    "producer_headroom": 3.5302708090429036,
+    "producer_batch_seconds": 0.08226992425001356,
+    "producer_p95_seconds": 0.20347445600054925,
+    "gpu_update_seconds": 0.9903872151375424,
+    "producer_batches_per_second": 12.155110255857993,
+    "gpu_batches_per_second": 1.009706087392417,
+    "producer_headroom": 12.038265795988977,
     "sustainable": true,
-    "buffer_seconds": 60.11907441102704,
+    "buffer_seconds": 60.41362012339008,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378330.52753664996,
-    "upload_bytes_per_second": 5685180.552910053,
+    "generation_bytes_per_second": 378790.75367362006,
+    "upload_bytes_per_second": 5763708.5439539505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.18092412542939,
+    "estimated_full_shards_per_hour": 81.279677940907,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 6680691,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0957601480004087,
+      "bytes": 10587203,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -83,3 +83,4 @@
 {"step": 363300, "loss": 0.6225243136286736, "lr": 0.0003, "grad_norm": 0.18326852934646368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53176756150, "tokens_per_second": 149920.7606125064, "data_wait_s": 1.2214000889798626e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 363400, "loss": 0.5955623667687178, "lr": 0.0003, "grad_norm": 0.15571062923710338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53191428154, "tokens_per_second": 142148.8241390262, "data_wait_s": 7.74100044509396e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 363500, "loss": 0.6045342851430178, "lr": 0.0003, "grad_norm": 0.1777388046768215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53206038919, "tokens_per_second": 149298.15606967517, "data_wait_s": 1.7934000425157137e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 363600, "loss": 0.5906962864100933, "lr": 0.0003, "grad_norm": 0.1517345435929581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53220693262, "tokens_per_second": 149365.92333611613, "data_wait_s": 1.32630011648871e-05, "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 sha256:007ae204df85e1904d0b88fd8f593dd73d9cd9b5b111864dc330fb342fe1b12c
-size 180978
+oid sha256:79dd91602c986ebfb2e982ffbb34f0172e66aeb955ce27e95a767113b2622aee
+size 183096