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": 75506.02289831277,
-      "required_pages_per_second_upper_estimate": 0.006544670696560289
+      "required_token_positions_per_second": 75431.9727805915,
+      "required_pages_per_second_upper_estimate": 0.006538252220034517
     },
     {
       "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": 9438.252862289097,
-      "required_pages_per_second_upper_estimate": 0.003765060181222713
+      "required_token_positions_per_second": 9428.996597573938,
+      "required_pages_per_second_upper_estimate": 0.0037613677188343456
     },
     {
       "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": 18876.505724578194,
-      "required_pages_per_second_upper_estimate": 0.004366747384226969
+      "required_token_positions_per_second": 18857.993195147876,
+      "required_pages_per_second_upper_estimate": 0.0043624648363526555
     },
     {
       "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": 28314.75858686729,
-      "required_pages_per_second_upper_estimate": 0.0017242202418829298
+      "required_token_positions_per_second": 28286.989792721815,
+      "required_pages_per_second_upper_estimate": 0.0017225292680110736
     },
     {
       "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": 28314.75858686729,
-      "required_pages_per_second_upper_estimate": 0.002179748496285373
+      "required_token_positions_per_second": 28286.989792721815,
+      "required_pages_per_second_upper_estimate": 0.002177610777643112
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28314.75858686729,
-      "required_pages_per_second_upper_estimate": 0.3515745382478524
+      "required_token_positions_per_second": 28286.989792721815,
+      "required_pages_per_second_upper_estimate": 0.3512297427607412
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14284253294556493,
-    "producer_p95_seconds": 0.3418100549970404,
-    "gpu_update_seconds": 0.9764519063504733,
-    "producer_batches_per_second": 7.000715958887991,
-    "gpu_batches_per_second": 1.024115978981022,
-    "producer_headroom": 6.83586244387436,
+    "producer_batch_seconds": 0.10490018369534937,
+    "producer_p95_seconds": 0.3396480869996594,
+    "gpu_update_seconds": 0.9774104704175265,
+    "producer_batches_per_second": 9.532871771742512,
+    "gpu_batches_per_second": 1.023111609979811,
+    "producer_headroom": 9.317528682848806,
     "sustainable": true,
-    "buffer_seconds": 60.540018193729345,
+    "buffer_seconds": 60.59944916588664,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316107.718365645,
-    "upload_bytes_per_second": 6110587.418275576,
+    "generation_bytes_per_second": 359580.94752402743,
+    "upload_bytes_per_second": 6000198.357483035,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.82935775019656,
+    "estimated_full_shards_per_hour": 77.15770072260491,
     "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": 1.150122408002062,
-      "bytes": 11429829,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.3652685859997291,
       "bytes": 10381816,
@@ -562,6 +556,12 @@
       "bytes": 5790107,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.9691603890023543,
+      "bytes": 2643836,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 111200, "loss": 0.6151546761393547, "lr": 0.0003, "grad_norm": 0.14512287742372176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16271827997, "tokens_per_second": 148151.61254821444, "data_wait_s": 7.84599978942424e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 111300, "loss": 0.6296935249119997, "lr": 0.0003, "grad_norm": 0.1266021879014762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16286494137, "tokens_per_second": 149323.95690671672, "data_wait_s": 1.8397993699181825e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 111400, "loss": 0.6208535227924585, "lr": 0.0003, "grad_norm": 0.142977013173407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16301263094, "tokens_per_second": 150214.72800092027, "data_wait_s": 1.4746001397725195e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 111500, "loss": 0.6298375278711319, "lr": 0.0003, "grad_norm": 0.1650104801879643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16315974137, "tokens_per_second": 128759.55248611214, "data_wait_s": 1.6564998077228665e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4caa2c79f75a8b731b7f6b766daf8db5192b6e716d1d86d306b25b9268aa97ce
-size 563232
+oid sha256:70f26c02c8ed29af7a8a9e5ab22c28b22a58db1898f7dab1a160533bb17d35c7
+size 565350