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": 77302.1223930104,
-      "required_pages_per_second_upper_estimate": 0.005347940192784174
+      "required_token_positions_per_second": 77307.73533094037,
+      "required_pages_per_second_upper_estimate": 0.005348328508853979
     },
     {
       "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": 9662.7652991263,
-      "required_pages_per_second_upper_estimate": 0.004045598049259399
+      "required_token_positions_per_second": 9663.466916367546,
+      "required_pages_per_second_upper_estimate": 0.004045891801746874
     },
     {
       "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": 19325.5305982526,
-      "required_pages_per_second_upper_estimate": 0.004024998375101762
+      "required_token_positions_per_second": 19326.933832735092,
+      "required_pages_per_second_upper_estimate": 0.004025290631838684
     },
     {
       "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": 28988.2958973789,
-      "required_pages_per_second_upper_estimate": 0.0017378040764829154
+      "required_token_positions_per_second": 28990.40074910264,
+      "required_pages_per_second_upper_estimate": 0.001737930259129831
     },
     {
       "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": 28988.2958973789,
-      "required_pages_per_second_upper_estimate": 0.0022537124999828105
+      "required_token_positions_per_second": 28990.40074910264,
+      "required_pages_per_second_upper_estimate": 0.002253876142946067
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28988.2958973789,
-      "required_pages_per_second_upper_estimate": 0.3718307345644476
+      "required_token_positions_per_second": 28990.40074910264,
+      "required_pages_per_second_upper_estimate": 0.371857733342346
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12196040357042648,
-    "producer_p95_seconds": 0.3336775379939354,
-    "gpu_update_seconds": 0.9537642398116928,
-    "producer_batches_per_second": 8.199382510427217,
-    "gpu_batches_per_second": 1.048477137492003,
-    "producer_headroom": 7.820277826982904,
+    "producer_batch_seconds": 0.29486860692998107,
+    "producer_p95_seconds": 0.34055492100014817,
+    "gpu_update_seconds": 0.953694991638079,
+    "producer_batches_per_second": 3.391341012566516,
+    "gpu_batches_per_second": 1.0485532678350202,
+    "producer_headroom": 3.234304938621498,
     "sustainable": true,
-    "buffer_seconds": 60.08714710813665,
+    "buffer_seconds": 60.08278447319898,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404920.6724565852,
-    "upload_bytes_per_second": 5596837.472966093,
+    "generation_bytes_per_second": 381860.0640729278,
+    "upload_bytes_per_second": 5636712.824501413,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.88655023835342,
+    "estimated_full_shards_per_hour": 81.93828050270916,
     "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": 0.8795082219985488,
-      "bytes": 997738,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.1275165429979097,
       "bytes": 11698586,
@@ -562,6 +556,12 @@
       "bytes": 6272554,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2908939679982723,
+      "bytes": 6343120,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -383,3 +383,4 @@
 {"step": 723300, "loss": 0.5442550778388977, "lr": 8e-05, "grad_norm": 0.21169117058640777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105871498737, "tokens_per_second": 152329.35673679033, "data_wait_s": 1.446500391466543e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723400, "loss": 0.5364526510238647, "lr": 8e-05, "grad_norm": 0.2332380704781859, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105885999701, "tokens_per_second": 151139.34921965763, "data_wait_s": 1.1629002983681858e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723500, "loss": 0.5505629181861877, "lr": 8e-05, "grad_norm": 0.23283509450779433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105900622546, "tokens_per_second": 152449.045320536, "data_wait_s": 1.4262004697229713e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 723600, "loss": 0.5499066710472107, "lr": 8e-05, "grad_norm": 0.21619974403789877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105915173698, "tokens_per_second": 151553.16999639873, "data_wait_s": 1.2883996532764286e-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 sha256:014ce36cf6017ea4a2bf23344970db8cd6e2861ddecbac5bb4484d8edbd8ba69
-size 821076
+oid sha256:40629c0374a28b166223364c104131fe0b74ae599d182974ae45a7511212f7e3
+size 823194