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": 75090.97369864154,
-      "required_pages_per_second_upper_estimate": 0.006508695283865365
+      "required_token_positions_per_second": 74996.83210355266,
+      "required_pages_per_second_upper_estimate": 0.006500535329002753
     },
     {
       "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": 9386.371712330192,
-      "required_pages_per_second_upper_estimate": 0.003744364014811789
+      "required_token_positions_per_second": 9374.604012944083,
+      "required_pages_per_second_upper_estimate": 0.003739669703583885
     },
     {
       "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": 18772.743424660384,
-      "required_pages_per_second_upper_estimate": 0.004342743802295091
+      "required_token_positions_per_second": 18749.208025888165,
+      "required_pages_per_second_upper_estimate": 0.00433729930199801
     },
     {
       "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": 28159.115136990575,
-      "required_pages_per_second_upper_estimate": 0.0017147423723835092
+      "required_token_positions_per_second": 28123.812038832246,
+      "required_pages_per_second_upper_estimate": 0.0017125925989267083
     },
     {
       "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": 28159.115136990575,
-      "required_pages_per_second_upper_estimate": 0.0021677666326652186
+      "required_token_positions_per_second": 28123.812038832246,
+      "required_pages_per_second_upper_estimate": 0.00216504890244376
     },
     {
       "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": 28159.115136990575,
-      "required_pages_per_second_upper_estimate": 0.3496419675055015
+      "required_token_positions_per_second": 28123.812038832246,
+      "required_pages_per_second_upper_estimate": 0.3492036211782441
     }
   ],
   "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": 0.0812333215467902,
-    "producer_p95_seconds": 0.21399625199956063,
-    "gpu_update_seconds": 0.981849034158067,
-    "producer_batches_per_second": 12.310219266658972,
-    "gpu_batches_per_second": 1.018486513924717,
-    "producer_headroom": 12.086776897243139,
+    "producer_batch_seconds": 0.13426086329687337,
+    "producer_p95_seconds": 0.29569887400066364,
+    "gpu_update_seconds": 0.9830815240062313,
+    "producer_batches_per_second": 7.448186876236835,
+    "gpu_batches_per_second": 1.0172096368211896,
+    "producer_headroom": 7.322174905374119,
     "sustainable": true,
-    "buffer_seconds": 60.874640117800155,
+    "buffer_seconds": 60.95105448838634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317131.77631160105,
-    "upload_bytes_per_second": 4067390.1755015077,
+    "generation_bytes_per_second": 328301.7625238092,
+    "upload_bytes_per_second": 4066234.791203969,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.04909674654984,
+    "estimated_full_shards_per_hour": 70.44591576371867,
     "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": 1.3130787809996036,
-      "bytes": 11524150,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.0262837310001487,
       "bytes": 2640097,
@@ -562,6 +556,12 @@
       "bytes": 7130727,
       "pages": 44,
       "shards": 2
+    },
+    {
+      "seconds": 1.1150904570004059,
+      "bytes": 10599538,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -138,3 +138,4 @@
 {"step": 328800, "loss": 0.660380631685257, "lr": 0.0003, "grad_norm": 0.20252718627519192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48127535874, "tokens_per_second": 148482.7347440647, "data_wait_s": 1.6338999557774514e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 328900, "loss": 0.6235296614468098, "lr": 0.0003, "grad_norm": 0.14611748385605103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48142190297, "tokens_per_second": 149227.0087026593, "data_wait_s": 1.6834999769343995e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 329000, "loss": 0.5691296383738518, "lr": 0.0003, "grad_norm": 0.17480108352651663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48156935805, "tokens_per_second": 149903.9100649539, "data_wait_s": 1.6370999219361693e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 329100, "loss": 0.5838132090866566, "lr": 0.0003, "grad_norm": 0.15181576133042385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48171621084, "tokens_per_second": 148798.20042659793, "data_wait_s": 1.4307999663287774e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33af4bca036f93aed74f457da69eea159aec4daffdd9e852bc603be2cda2c1d7
-size 298251
+oid sha256:521bb7b2ec5f5b63c2ab23f359ab22f0ab2757c36f094f6ded4042b935a2c4e4
+size 300369