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": 74541.83390865808,
-      "required_pages_per_second_upper_estimate": 0.006461097238651673
+      "required_token_positions_per_second": 73286.81074761767,
+      "required_pages_per_second_upper_estimate": 0.006352315011879811
     },
     {
       "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": 9317.72923858226,
-      "required_pages_per_second_upper_estimate": 0.0037169815057372987
+      "required_token_positions_per_second": 9160.851343452208,
+      "required_pages_per_second_upper_estimate": 0.0036544005678363686
     },
     {
       "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": 18635.45847716452,
-      "required_pages_per_second_upper_estimate": 0.004310985345824478
+      "required_token_positions_per_second": 18321.702686904417,
+      "required_pages_per_second_upper_estimate": 0.004238403465661115
     },
     {
       "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": 27953.187715746782,
-      "required_pages_per_second_upper_estimate": 0.0017022024728474425
+      "required_token_positions_per_second": 27482.554030356623,
+      "required_pages_per_second_upper_estimate": 0.0016735433506312972
     },
     {
       "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": 27953.187715746782,
-      "required_pages_per_second_upper_estimate": 0.002151913769734285
+      "required_token_positions_per_second": 27482.554030356623,
+      "required_pages_per_second_upper_estimate": 0.0021156830858355268
     },
     {
       "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": 27953.187715746782,
-      "required_pages_per_second_upper_estimate": 0.3470850381283979
+      "required_token_positions_per_second": 27482.554030356623,
+      "required_pages_per_second_upper_estimate": 0.3412413428654733
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5183917470938013,
-    "producer_p95_seconds": 0.35891752800307586,
-    "gpu_update_seconds": 0.989082185586481,
-    "producer_batches_per_second": 1.9290430559633374,
-    "gpu_batches_per_second": 1.011038328839221,
-    "producer_headroom": 1.907982121882642,
+    "producer_batch_seconds": 0.43211702229669413,
+    "producer_p95_seconds": 0.33367446499687503,
+    "gpu_update_seconds": 1.0060200361822496,
+    "producer_batches_per_second": 2.314187936140581,
+    "gpu_batches_per_second": 0.994015987787783,
+    "producer_headroom": 2.328119431248673,
     "sustainable": true,
-    "buffer_seconds": 60.33401332077534,
+    "buffer_seconds": 60.36120217093497,
     "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": 289171.2533846425,
-    "upload_bytes_per_second": 3770843.6606367407,
+    "generation_bytes_per_second": 266264.8417269639,
+    "upload_bytes_per_second": 3764138.0954640657,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.04941941408592,
+    "estimated_full_shards_per_hour": 57.13423670632064,
     "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": 0.9710190369987686,
-      "bytes": 6743419,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.0164730639990012,
       "bytes": 6038957,
@@ -562,6 +556,12 @@
       "bytes": 7358899,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.2355712200005655,
+      "bytes": 6983067,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -297,3 +297,4 @@
 {"step": 69700, "loss": 0.6128690466284752, "lr": 0.0003, "grad_norm": 0.1299160860011532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10201051763, "tokens_per_second": 150101.95188452434, "data_wait_s": 1.0831001418409869e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 69800, "loss": 0.6174406055361032, "lr": 0.0003, "grad_norm": 0.14424695000936255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10215705395, "tokens_per_second": 143158.04213653147, "data_wait_s": 8.422000973951072e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 69900, "loss": 0.6463001444935799, "lr": 0.0003, "grad_norm": 0.130460975222244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10230311326, "tokens_per_second": 130485.51989099575, "data_wait_s": 1.857100141933188e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 70000, "loss": 0.6447709798812866, "lr": 0.0003, "grad_norm": 0.13342261601705638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10244777274, "tokens_per_second": 147803.9612738101, "data_wait_s": 6.407000910257921e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1db41650f4a06a19abb98ca25d9fd8801228bd2578ded5a4a8779b62b3415e51
-size 637362
+oid sha256:99eea5c24692678f9652834160661dea62ea8520052da3a51cd85c8752253f50
+size 639480