Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76261.21962577773,
-      "required_pages_per_second_upper_estimate": 0.005275928124119844
+      "required_token_positions_per_second": 75807.6632220059,
+      "required_pages_per_second_upper_estimate": 0.00524455003971106
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9532.652453222216,
-      "required_pages_per_second_upper_estimate": 0.003991122517744548
+      "required_token_positions_per_second": 9475.957902750737,
+      "required_pages_per_second_upper_estimate": 0.003967385693378982
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19065.30490644443,
-      "required_pages_per_second_upper_estimate": 0.003970800226111105
+      "required_token_positions_per_second": 18951.915805501474,
+      "required_pages_per_second_upper_estimate": 0.003947184266656366
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28597.957359666645,
-      "required_pages_per_second_upper_estimate": 0.0017144038771595026
+      "required_token_positions_per_second": 28427.873708252213,
+      "required_pages_per_second_upper_estimate": 0.0017042076219599058
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28597.957359666645,
-      "required_pages_per_second_upper_estimate": 0.0022233653955934605
+      "required_token_positions_per_second": 28427.873708252213,
+      "required_pages_per_second_upper_estimate": 0.0022101421398149075
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28597.957359666645,
-      "required_pages_per_second_upper_estimate": 0.36682389091554296
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28427.873708252213,
+      "required_pages_per_second_upper_estimate": 0.3646422404567952
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11464789210941717,
-    "producer_p95_seconds": 0.33563471299567027,
-    "gpu_update_seconds": 0.9667823352654401,
-    "producer_batches_per_second": 8.722358358282106,
-    "gpu_batches_per_second": 1.03435899014998,
-    "producer_headroom": 8.432621982642004,
+    "producer_batch_seconds": 0.14001098451558391,
+    "producer_p95_seconds": 0.33739263899769867,
+    "gpu_update_seconds": 0.9725665831973277,
+    "producer_batches_per_second": 7.142296752357277,
+    "gpu_batches_per_second": 1.0282072377116687,
+    "producer_headroom": 6.946359148621487,
     "sustainable": true,
-    "buffer_seconds": 60.907287121722725,
+    "buffer_seconds": 60.29912815823432,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438304.62170390435,
-    "upload_bytes_per_second": 5641860.546302463,
+    "generation_bytes_per_second": 421480.6329628248,
+    "upload_bytes_per_second": 5625618.4384005545,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.04996860826347,
+    "estimated_full_shards_per_hour": 90.43993226684148,
     "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": 0.9289632090003579,
-      "bytes": 5852982,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0356132939996314,
       "bytes": 7513178,
@@ -562,6 +556,12 @@
       "bytes": 12455748,
       "pages": 43,
       "shards": 4
+    },
+    {
+      "seconds": 1.187473874000716,
+      "bytes": 6085388,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -375,3 +375,4 @@
 {"step": 722500, "loss": 0.519875705242157, "lr": 8e-05, "grad_norm": 0.22840929185045208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105754059450, "tokens_per_second": 151841.19993489745, "data_wait_s": 1.5067998901940882e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 722600, "loss": 0.5492111444473267, "lr": 8e-05, "grad_norm": 0.22936311170345755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105768734296, "tokens_per_second": 152930.9104988477, "data_wait_s": 1.4805998944211751e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 722700, "loss": 0.5891305804252625, "lr": 8e-05, "grad_norm": 0.23494695261280726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105783284226, "tokens_per_second": 151962.07370628853, "data_wait_s": 9.780000254977494e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 722800, "loss": 0.5541865229606628, "lr": 8e-05, "grad_norm": 0.21566273233327088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105798063516, "tokens_per_second": 154729.68355813564, "data_wait_s": 1.4995996025390923e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2af9e8e6c680f9cadffc7076a1635de288e1bf9db0fd3c9188a844e0fde7d19a
-size 804132
+oid sha256:4d0c64575c6a22b0812fb1e1758a78493fc6de11e432f4157c1e2679f54b7701
+size 806250