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": 74297.3885843981,
-      "required_pages_per_second_upper_estimate": 0.006439909337485828
+      "required_token_positions_per_second": 74898.20123207444,
+      "required_pages_per_second_upper_estimate": 0.0064919862550406795
     },
     {
       "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": 9287.173573049762,
-      "required_pages_per_second_upper_estimate": 0.0037047923939084737
+      "required_token_positions_per_second": 9362.275154009305,
+      "required_pages_per_second_upper_estimate": 0.0037347515374219343
     },
     {
       "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": 18574.347146099524,
-      "required_pages_per_second_upper_estimate": 0.004296848314905284
+      "required_token_positions_per_second": 18724.55030801861,
+      "required_pages_per_second_upper_estimate": 0.004331595172929901
     },
     {
       "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": 27861.52071914929,
-      "required_pages_per_second_upper_estimate": 0.0016966204336942173
+      "required_token_positions_per_second": 28086.825462027915,
+      "required_pages_per_second_upper_estimate": 0.001710340310452896
     },
     {
       "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": 27861.52071914929,
-      "required_pages_per_second_upper_estimate": 0.0021448569905857225
+      "required_token_positions_per_second": 28086.825462027915,
+      "required_pages_per_second_upper_estimate": 0.002162201573375965
     },
     {
       "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": 27861.52071914929,
-      "required_pages_per_second_upper_estimate": 0.3459468408203594
+      "required_token_positions_per_second": 28086.825462027915,
+      "required_pages_per_second_upper_estimate": 0.348744371680444
     }
   ],
   "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": 0.12615877163284495,
-    "producer_p95_seconds": 0.40894824999850243,
-    "gpu_update_seconds": 0.992336358043711,
-    "producer_batches_per_second": 7.926519789763503,
-    "gpu_batches_per_second": 1.0077228269368232,
-    "producer_headroom": 7.865773780135317,
+    "producer_batch_seconds": 0.4681365180549051,
+    "producer_p95_seconds": 0.4173383509987616,
+    "gpu_update_seconds": 0.984376110336101,
+    "producer_batches_per_second": 2.136129016712846,
+    "gpu_batches_per_second": 1.0158718700096903,
+    "producer_headroom": 2.1027543726478717,
     "sustainable": true,
-    "buffer_seconds": 60.53251784066637,
+    "buffer_seconds": 60.04694273050216,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318563.6002777978,
-    "upload_bytes_per_second": 5857817.227429982,
+    "generation_bytes_per_second": 293805.1047517916,
+    "upload_bytes_per_second": 5871650.838530905,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.3563328385396,
+    "estimated_full_shards_per_hour": 63.043736046937084,
     "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": 0.9573843099969963,
-      "bytes": 845578,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.3486182489978091,
       "bytes": 12484307,
@@ -562,6 +556,12 @@
       "bytes": 12004299,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.7972624179965351,
+      "bytes": 919254,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -384,3 +384,4 @@
 {"step": 233400, "loss": 0.5881534330546856, "lr": 0.0003, "grad_norm": 0.15923444566418937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34163539739, "tokens_per_second": 151573.37001466786, "data_wait_s": 9.046998457051814e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 233500, "loss": 0.4985401816666126, "lr": 0.0003, "grad_norm": 0.13233641011653927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34178233929, "tokens_per_second": 147326.7281481473, "data_wait_s": 1.691799843683839e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 233600, "loss": 0.6095574572682381, "lr": 0.0003, "grad_norm": 0.17349696892607225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34192695865, "tokens_per_second": 146619.6809009955, "data_wait_s": 1.079399953596294e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 233700, "loss": 0.5399864874780178, "lr": 0.0003, "grad_norm": 0.1503727513628934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34207228558, "tokens_per_second": 147259.34759513917, "data_wait_s": 8.631999662611634e-06, "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:fa13f4c49294ac6753555ce55b5db274e5d59a9946b433dcb1fb19ae86edd2fa
-size 823194
+oid sha256:8da08318c07cf79b2d05c30065eac39d93bd3a44c7de9426e416737b12c9c876
+size 825312