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": 74957.95188930034,
-      "required_pages_per_second_upper_estimate": 0.006497165290572364
+      "required_token_positions_per_second": 75139.92167780774,
+      "required_pages_per_second_upper_estimate": 0.006512937970641973
     },
     {
       "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": 9369.743986162543,
-      "required_pages_per_second_upper_estimate": 0.003737730966236853
+      "required_token_positions_per_second": 9392.490209725967,
+      "required_pages_per_second_upper_estimate": 0.003746804774902652
     },
     {
       "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": 18739.487972325085,
-      "required_pages_per_second_upper_estimate": 0.004335050738673291
+      "required_token_positions_per_second": 18784.980419451935,
+      "required_pages_per_second_upper_estimate": 0.004345574615676364
     },
     {
       "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": 28109.23195848763,
-      "required_pages_per_second_upper_estimate": 0.001711704748529491
+      "required_token_positions_per_second": 28177.470629177904,
+      "required_pages_per_second_upper_estimate": 0.0017158601255538928
     },
     {
       "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": 28109.23195848763,
-      "required_pages_per_second_upper_estimate": 0.002163926487498586
+      "required_token_positions_per_second": 28177.470629177904,
+      "required_pages_per_second_upper_estimate": 0.0021691796892650537
     },
     {
       "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": 28109.23195848763,
-      "required_pages_per_second_upper_estimate": 0.34902258537675396
+      "required_token_positions_per_second": 28177.470629177904,
+      "required_pages_per_second_upper_estimate": 0.3498698812865876
     }
   ],
   "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": 0.06487477017182641,
-    "producer_p95_seconds": 0.2090967020012613,
-    "gpu_update_seconds": 0.9835914421579078,
-    "producer_batches_per_second": 15.414312796044038,
-    "gpu_batches_per_second": 1.0166822901652066,
-    "producer_headroom": 15.16138615293405,
+    "producer_batch_seconds": 0.2983639846171826,
+    "producer_p95_seconds": 0.2563801289998082,
+    "gpu_update_seconds": 0.9812094337300228,
+    "producer_batches_per_second": 3.3516109569425914,
+    "gpu_batches_per_second": 1.019150413381724,
+    "producer_headroom": 3.2886322891449797,
     "sustainable": true,
-    "buffer_seconds": 60.98266941379028,
+    "buffer_seconds": 60.83498489126141,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405600.07525056845,
-    "upload_bytes_per_second": 5308221.154932217,
+    "generation_bytes_per_second": 450951.7568081669,
+    "upload_bytes_per_second": 5229687.106816401,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.03233426225462,
+    "estimated_full_shards_per_hour": 96.76374939140086,
     "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": 1.2374535429999014,
-      "bytes": 7166128,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.0863462409997737,
       "bytes": 6279325,
@@ -562,6 +556,12 @@
       "bytes": 6092850,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.532518773999982,
+      "bytes": 2518854,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -112,3 +112,4 @@
 {"step": 206200, "loss": 0.5746875181794167, "lr": 0.0003, "grad_norm": 0.14248435960817618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30179754206, "tokens_per_second": 147307.53043465337, "data_wait_s": 1.621200135559775e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 206300, "loss": 0.5879179909825325, "lr": 0.0003, "grad_norm": 0.13954754362615665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30194524871, "tokens_per_second": 149234.82555407833, "data_wait_s": 1.6312000298057683e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 206400, "loss": 0.6384740583598614, "lr": 0.0003, "grad_norm": 0.17451350463504176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30209255939, "tokens_per_second": 148678.38793833062, "data_wait_s": 2.5233000997104682e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 206500, "loss": 0.5908651724457741, "lr": 0.0003, "grad_norm": 0.1655434847203432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30223869542, "tokens_per_second": 147681.66956276508, "data_wait_s": 1.7620999642531388e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1c72f7777e37ed02b2c32c0530514a44e655cf1544efd6968c02346e8ba7032
-size 243183
+oid sha256:34fac9a54bc856d0be8aee566f96f5a7702fbd424ddbbc985599b8ab10530368
+size 245301