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": 77125.60200302312,
-      "required_pages_per_second_upper_estimate": 0.005335728102621119
+      "required_token_positions_per_second": 77136.00753868619,
+      "required_pages_per_second_upper_estimate": 0.005336447981722459
     },
     {
       "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": 9640.70025037789,
-      "required_pages_per_second_upper_estimate": 0.004036359874119053
+      "required_token_positions_per_second": 9642.000942335773,
+      "required_pages_per_second_upper_estimate": 0.004036904446680282
     },
     {
       "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": 19281.40050075578,
-      "required_pages_per_second_upper_estimate": 0.004015807239582107
+      "required_token_positions_per_second": 19284.001884671547,
+      "required_pages_per_second_upper_estimate": 0.0040163490392486545
     },
     {
       "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": 28922.10075113367,
-      "required_pages_per_second_upper_estimate": 0.0017338357785396496
+      "required_token_positions_per_second": 28926.002827007316,
+      "required_pages_per_second_upper_estimate": 0.00173406970203015
     },
     {
       "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": 28922.10075113367,
-      "required_pages_per_second_upper_estimate": 0.002248566118523921
+      "required_token_positions_per_second": 28926.002827007316,
+      "required_pages_per_second_upper_estimate": 0.002248869487759679
     },
     {
       "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": 28922.10075113367,
-      "required_pages_per_second_upper_estimate": 0.3709816543032243
+      "required_token_positions_per_second": 28926.002827007316,
+      "required_pages_per_second_upper_estimate": 0.37103170594280876
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07217460371099094,
-    "producer_p95_seconds": 0.19134322799982328,
-    "gpu_update_seconds": 0.9559471574317184,
-    "producer_batches_per_second": 13.85528909870159,
-    "gpu_batches_per_second": 1.0460829264732954,
-    "producer_headroom": 13.24492422929846,
+    "producer_batch_seconds": 0.08842671753114928,
+    "producer_p95_seconds": 0.17762938499981828,
+    "gpu_update_seconds": 0.955818201545148,
+    "producer_batches_per_second": 11.308799285099994,
+    "gpu_batches_per_second": 1.0462240605833086,
+    "producer_headroom": 10.80915619431933,
     "sustainable": true,
-    "buffer_seconds": 60.224670918198264,
+    "buffer_seconds": 60.21654669734432,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367530.8485784619,
-    "upload_bytes_per_second": 5667298.512563198,
+    "generation_bytes_per_second": 366903.04132651945,
+    "upload_bytes_per_second": 5622915.661866501,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.86356442466158,
+    "estimated_full_shards_per_hour": 78.72885160299957,
     "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": 1.1055204159999903,
-      "bytes": 6016330,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9128079020001678,
       "bytes": 2402677,
@@ -562,6 +556,12 @@
       "bytes": 11149283,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.0994711280000047,
+      "bytes": 2636989,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -84,3 +84,4 @@
 {"step": 565500, "loss": 0.5776864290237427, "lr": 0.0003, "grad_norm": 0.14738988748703424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82771885314, "tokens_per_second": 151557.58958653844, "data_wait_s": 1.4744000509381294e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 565600, "loss": 0.7269312739372253, "lr": 0.0003, "grad_norm": 0.231847690184574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82786542230, "tokens_per_second": 152578.4501092566, "data_wait_s": 1.3545000911108218e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 565700, "loss": 0.5631712079048157, "lr": 0.0003, "grad_norm": 0.15288066367707487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82801274464, "tokens_per_second": 153648.1767180552, "data_wait_s": 9.246999979950488e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 565800, "loss": 0.5782396197319031, "lr": 0.0003, "grad_norm": 0.16403070827311558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82815974215, "tokens_per_second": 153332.08820165443, "data_wait_s": 1.2675000107265078e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba73599ed3cc24adcbf3589165668e40a1d2fe4745f0cc69c386363147993ca0
-size 183879
+oid sha256:4af15aebf8cd92ca218192ca1ada9b3197fb3aa26c1159a7f05cf04d26610391
+size 185997