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": 74825.75348463499,
-      "required_pages_per_second_upper_estimate": 0.006485706667909765
+      "required_token_positions_per_second": 74805.53007602457,
+      "required_pages_per_second_upper_estimate": 0.006483953754107175
     },
     {
       "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": 9353.219185579374,
-      "required_pages_per_second_upper_estimate": 0.0037311389762164407
+      "required_token_positions_per_second": 9350.691259503072,
+      "required_pages_per_second_upper_estimate": 0.003730130548708741
     },
     {
       "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": 18706.438371158747,
-      "required_pages_per_second_upper_estimate": 0.004327405295096573
+      "required_token_positions_per_second": 18701.382519006143,
+      "required_pages_per_second_upper_estimate": 0.004326235712681027
     },
     {
       "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": 28059.657556738122,
-      "required_pages_per_second_upper_estimate": 0.0017086859275597306
+      "required_token_positions_per_second": 28052.073778509213,
+      "required_pages_per_second_upper_estimate": 0.0017082241152545994
     },
     {
       "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": 28059.657556738122,
-      "required_pages_per_second_upper_estimate": 0.0021601101128211815
+      "required_token_positions_per_second": 28052.073778509213,
+      "required_pages_per_second_upper_estimate": 0.0021595262925798784
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28059.657556738122,
-      "required_pages_per_second_upper_estimate": 0.3484070372218747
+      "required_token_positions_per_second": 28052.073778509213,
+      "required_pages_per_second_upper_estimate": 0.34831287207754463
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.20454988270324748,
-    "producer_p95_seconds": 0.348550978997082,
-    "gpu_update_seconds": 0.9853292024000746,
-    "producer_batches_per_second": 4.888783052741999,
-    "gpu_batches_per_second": 1.0148892345463731,
-    "producer_headroom": 4.817060706065275,
+    "producer_batch_seconds": 0.362052130859297,
+    "producer_p95_seconds": 0.33937483499903465,
+    "gpu_update_seconds": 0.9855955826403545,
+    "producer_batches_per_second": 2.7620331846317083,
+    "gpu_batches_per_second": 1.014614937012052,
+    "producer_headroom": 2.722247705879082,
     "sustainable": true,
-    "buffer_seconds": 60.10508134640455,
+    "buffer_seconds": 60.121330541061624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448366.07144974644,
-    "upload_bytes_per_second": 5517080.547400296,
+    "generation_bytes_per_second": 407227.6649371979,
+    "upload_bytes_per_second": 5564294.956621541,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.20892150515837,
+    "estimated_full_shards_per_hour": 87.38157712065652,
     "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": 1.0467319720009982,
-      "bytes": 5732511,
-      "pages": 41,
-      "shards": 2
-    },
     {
       "seconds": 1.2224297300017497,
       "bytes": 7561441,
@@ -562,6 +556,12 @@
       "bytes": 11781038,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.124476634999155,
+      "bytes": 9801537,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -258,3 +258,4 @@
 {"step": 180800, "loss": 0.5887566804885864, "lr": 0.0003, "grad_norm": 0.1595766333864267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26461442479, "tokens_per_second": 149108.69555331505, "data_wait_s": 1.1955999070778489e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 180900, "loss": 0.6022106818854809, "lr": 0.0003, "grad_norm": 0.1460082143141704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26476208702, "tokens_per_second": 149031.7166816439, "data_wait_s": 1.2014999811071903e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 181000, "loss": 0.5989765413105488, "lr": 0.0003, "grad_norm": 0.17027742300177381, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26490912366, "tokens_per_second": 149997.79116370252, "data_wait_s": 1.3525001122616231e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 181100, "loss": 0.6466478742659092, "lr": 0.0003, "grad_norm": 0.1611418512595973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26505398020, "tokens_per_second": 147473.10472566748, "data_wait_s": 1.1824999091913924e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f583c226e83f86ab8e3826e9b44ad7ae186d91103c3c84e1fa0d600be8165ac
-size 554760
+oid sha256:ae918582456c9fcb2869453f8edf685982569f27c062e7037de6cd8f942baf7f
+size 556878