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": 75616.69017530579,
-      "required_pages_per_second_upper_estimate": 0.006554263055646395
+      "required_token_positions_per_second": 75641.90267512431,
+      "required_pages_per_second_upper_estimate": 0.006556448411230167
     },
     {
       "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": 9452.086271913224,
-      "required_pages_per_second_upper_estimate": 0.0037705785351496825
+      "required_token_positions_per_second": 9455.237834390538,
+      "required_pages_per_second_upper_estimate": 0.0037718357405419414
     },
     {
       "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": 18904.172543826448,
-      "required_pages_per_second_upper_estimate": 0.004373147615940664
+      "required_token_positions_per_second": 18910.475668781077,
+      "required_pages_per_second_upper_estimate": 0.0043746057329677075
     },
     {
       "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": 28356.258815739668,
-      "required_pages_per_second_upper_estimate": 0.001726747388086384
+      "required_token_positions_per_second": 28365.713503171613,
+      "required_pages_per_second_upper_estimate": 0.0017273231289460782
     },
     {
       "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": 28356.258815739668,
-      "required_pages_per_second_upper_estimate": 0.0021829432987839547
+      "required_token_positions_per_second": 28365.713503171613,
+      "required_pages_per_second_upper_estimate": 0.0021836711467947156
     },
     {
       "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": 28356.258815739668,
-      "required_pages_per_second_upper_estimate": 0.352089832198116
+      "required_token_positions_per_second": 28365.713503171613,
+      "required_pages_per_second_upper_estimate": 0.35220722777321745
     }
   ],
   "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": 0.36957186944525233,
+    "producer_batch_seconds": 0.09075382133593735,
     "producer_p95_seconds": 0.20575111099969945,
-    "gpu_update_seconds": 0.9750228399189763,
-    "producer_batches_per_second": 2.705833648813842,
-    "gpu_batches_per_second": 1.025616999990584,
-    "producer_headroom": 2.638249608614798,
+    "gpu_update_seconds": 0.9746978512248118,
+    "producer_batches_per_second": 11.018819761851868,
+    "gpu_batches_per_second": 1.0259589664052233,
+    "producer_headroom": 10.740019944910507,
     "sustainable": true,
-    "buffer_seconds": 60.45141607497653,
+    "buffer_seconds": 60.43126677593833,
     "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": 276623.860689617,
-    "upload_bytes_per_second": 4063181.2369461562,
+    "generation_bytes_per_second": 277479.04848311335,
+    "upload_bytes_per_second": 4093432.6421139585,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.35704103008635,
+    "estimated_full_shards_per_hour": 59.54054442281771,
     "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": 1.0517817119998654,
-      "bytes": 7714712,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.9509194960000968,
       "bytes": 6981450,
@@ -562,6 +556,12 @@
       "bytes": 6539230,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.2508507479997206,
+      "bytes": 11653692,
+      "pages": 45,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -100,3 +100,4 @@
 {"step": 325000, "loss": 0.6160995494574308, "lr": 0.0003, "grad_norm": 0.14224377468636584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47571096641, "tokens_per_second": 151367.18161095746, "data_wait_s": 1.3118000424583443e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325100, "loss": 0.6068964377045631, "lr": 0.0003, "grad_norm": 0.1606168216520234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47585915270, "tokens_per_second": 134267.51756507542, "data_wait_s": 1.288900057261344e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325200, "loss": 0.5771619603037834, "lr": 0.0003, "grad_norm": 0.14875200384124426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47600513469, "tokens_per_second": 148641.97463967348, "data_wait_s": 1.4517001545755193e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 325300, "loss": 0.5691273473203182, "lr": 0.0003, "grad_norm": 0.1640405280227942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47614986024, "tokens_per_second": 147575.77974620298, "data_wait_s": 1.2975000572623685e-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 sha256:1c3abbcc51610177c75fd319f5271883fb5d371aedca47f823509052db206a53
-size 217767
+oid sha256:e3b72bf6447d719ba06d25a418b645577976a5dfac8f2e9278f9a620563c29f2
+size 219885