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": 76047.05839358868,
-      "required_pages_per_second_upper_estimate": 0.006591566282048878
+      "required_token_positions_per_second": 75766.67656300655,
+      "required_pages_per_second_upper_estimate": 0.0065672634955952726
     },
     {
       "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": 9505.882299198585,
-      "required_pages_per_second_upper_estimate": 0.0037920385747560975
+      "required_token_positions_per_second": 9470.834570375819,
+      "required_pages_per_second_upper_estimate": 0.0037780575117184532
     },
     {
       "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": 19011.76459839717,
-      "required_pages_per_second_upper_estimate": 0.004398037144210113
+      "required_token_positions_per_second": 18941.669140751637,
+      "required_pages_per_second_upper_estimate": 0.0043818217937217395
     },
     {
       "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": 28517.646897595754,
-      "required_pages_per_second_upper_estimate": 0.0017365750755335937
+      "required_token_positions_per_second": 28412.503711127458,
+      "required_pages_per_second_upper_estimate": 0.0017301724071213404
     },
     {
       "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": 28517.646897595754,
-      "required_pages_per_second_upper_estimate": 0.00219536740007604
+      "required_token_positions_per_second": 28412.503711127458,
+      "required_pages_per_second_upper_estimate": 0.002187273186526742
     },
     {
       "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": 28517.646897595754,
-      "required_pages_per_second_upper_estimate": 0.3540937320436042
+      "required_token_positions_per_second": 28412.503711127458,
+      "required_pages_per_second_upper_estimate": 0.3527882055592766
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.6594374040626008,
-    "producer_p95_seconds": 0.3815345389957656,
-    "gpu_update_seconds": 0.9695049559762565,
-    "producer_batches_per_second": 1.5164441595810199,
-    "gpu_batches_per_second": 1.0314542425345687,
-    "producer_headroom": 1.4702001281750479,
+    "producer_batch_seconds": 0.12457593414063695,
+    "producer_p95_seconds": 0.3720379350052099,
+    "gpu_update_seconds": 0.9730927017590482,
+    "producer_batches_per_second": 8.027232602334529,
+    "gpu_batches_per_second": 1.0276513205702928,
+    "producer_headroom": 7.8112414606540215,
     "sustainable": true,
-    "buffer_seconds": 60.10930727052791,
+    "buffer_seconds": 60.33174750906099,
     "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": 340899.7358401229,
-    "upload_bytes_per_second": 4205555.693378444,
+    "generation_bytes_per_second": 341120.0589091122,
+    "upload_bytes_per_second": 3824891.637189163,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.14914757159009,
+    "estimated_full_shards_per_hour": 73.19642377333663,
     "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": 1.2165515029992093,
-      "bytes": 11904873,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 0.9436809830003767,
       "bytes": 889392,
@@ -562,6 +556,12 @@
       "bytes": 6824656,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 11.391802897000161,
+      "bytes": 12565360,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -351,3 +351,4 @@
 {"step": 35100, "loss": 0.6899380572140217, "lr": 0.0003, "grad_norm": 0.15498536011013583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5136085480, "tokens_per_second": 122253.15101073103, "data_wait_s": 1.4624005416408181e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 35200, "loss": 0.6399045661091805, "lr": 0.0003, "grad_norm": 0.1344166233275482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5150764286, "tokens_per_second": 150483.99047347528, "data_wait_s": 8.450995665043592e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 35300, "loss": 0.6423078514635563, "lr": 0.0003, "grad_norm": 0.13669549508507228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5165289678, "tokens_per_second": 145017.1141338006, "data_wait_s": 1.1691998224705458e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 35400, "loss": 0.6142779737710953, "lr": 0.0003, "grad_norm": 0.13293723371427815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5180032624, "tokens_per_second": 149743.53608937512, "data_wait_s": 1.138600055128336e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f08589348b3a47bfb2702d3a9eecc33e948cf8784d0e646a688cc08f46458981
-size 747526
+oid sha256:89217a33fda14d6fd60964476bb8362d8a68cd12ecb981a5f9119dd85e8d19e0
+size 749644