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": 77043.06728792394,
-      "required_pages_per_second_upper_estimate": 0.0053300181595236345
+      "required_token_positions_per_second": 77008.47653304035,
+      "required_pages_per_second_upper_estimate": 0.005327625090839178
     },
     {
       "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": 9630.383410990493,
-      "required_pages_per_second_upper_estimate": 0.004032040428907655
+      "required_token_positions_per_second": 9626.059566630043,
+      "required_pages_per_second_upper_estimate": 0.004030230125566072
     },
     {
       "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": 19260.766821980986,
-      "required_pages_per_second_upper_estimate": 0.004011509788440019
+      "required_token_positions_per_second": 19252.119133260087,
+      "required_pages_per_second_upper_estimate": 0.004009708702934343
     },
     {
       "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": 28891.15023297148,
-      "required_pages_per_second_upper_estimate": 0.0017319803422345312
+      "required_token_positions_per_second": 28878.17869989013,
+      "required_pages_per_second_upper_estimate": 0.0017312027186326886
     },
     {
       "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": 28891.15023297148,
-      "required_pages_per_second_upper_estimate": 0.0022461598518737527
+      "required_token_positions_per_second": 28878.17869989013,
+      "required_pages_per_second_upper_estimate": 0.0022451513722324223
     },
     {
       "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": 28891.15023297148,
-      "required_pages_per_second_upper_estimate": 0.37058465428831694
+      "required_token_positions_per_second": 28878.17869989013,
+      "required_pages_per_second_upper_estimate": 0.3704182693896965
     }
   ],
   "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": 0.12773813970329684,
-    "producer_p95_seconds": 0.3074475980029092,
-    "gpu_update_seconds": 0.9569712447255646,
-    "producer_batches_per_second": 7.828515448265845,
-    "gpu_batches_per_second": 1.0449634777550447,
-    "producer_headroom": 7.491664172880277,
+    "producer_batch_seconds": 0.10904455229686505,
+    "producer_p95_seconds": 0.29280203099915525,
+    "gpu_update_seconds": 0.9574010981553068,
+    "producer_batches_per_second": 9.17056358099926,
+    "gpu_batches_per_second": 1.0444943106152391,
+    "producer_headroom": 8.779907643151756,
     "sustainable": true,
-    "buffer_seconds": 60.28918841771057,
+    "buffer_seconds": 60.31626918378433,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 413297.7938980088,
-    "upload_bytes_per_second": 5409450.973229825,
+    "generation_bytes_per_second": 412777.7590896337,
+    "upload_bytes_per_second": 5416572.012920619,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.68408549027632,
+    "estimated_full_shards_per_hour": 88.57249812618979,
     "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": 1.3453053759985778,
-      "bytes": 7563007,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.2517445409976062,
       "bytes": 10940783,
@@ -562,6 +556,12 @@
       "bytes": 16111480,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.2388989300015965,
+      "bytes": 7565588,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -283,3 +283,4 @@
 {"step": 673300, "loss": 0.5706678628921509, "lr": 0.0001, "grad_norm": 0.19494510413256247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98553506449, "tokens_per_second": 152480.47147365767, "data_wait_s": 1.3478998880600557e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673400, "loss": 0.5718695521354675, "lr": 0.0001, "grad_norm": 0.23013544544482514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98568308512, "tokens_per_second": 154093.95665601574, "data_wait_s": 1.2325999705353752e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673500, "loss": 0.6256731748580933, "lr": 0.0001, "grad_norm": 0.20255019155449147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98582989496, "tokens_per_second": 152949.60079562184, "data_wait_s": 1.3421999028651044e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 673600, "loss": 0.5435763001441956, "lr": 0.0001, "grad_norm": 0.2397423740731034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98597483724, "tokens_per_second": 150260.20550584595, "data_wait_s": 1.4335000741994008e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a92e625cfcdde7fd32958d55c2332709afef5cbb05b2cc36ded650e11fad697
-size 607710
+oid sha256:b02f1cd5ce5c3e8ea9e77822d50dc198beb70524ceb006bb79dbb6bc5e6f5790
+size 609828