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": 76645.97869466315,
-      "required_pages_per_second_upper_estimate": 0.005302546649269376
+      "required_token_positions_per_second": 77077.45183953545,
+      "required_pages_per_second_upper_estimate": 0.00533239696258727
     },
     {
       "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": 9580.747336832894,
-      "required_pages_per_second_upper_estimate": 0.004011258841176963
+      "required_token_positions_per_second": 9634.681479941932,
+      "required_pages_per_second_upper_estimate": 0.004033839940623736
     },
     {
       "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": 19161.494673665788,
-      "required_pages_per_second_upper_estimate": 0.003990834017928566
+      "required_token_positions_per_second": 19269.362959883863,
+      "required_pages_per_second_upper_estimate": 0.004013300137269788
     },
     {
       "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": 28742.242010498678,
-      "required_pages_per_second_upper_estimate": 0.0017230535216669774
+      "required_token_positions_per_second": 28904.044439825793,
+      "required_pages_per_second_upper_estimate": 0.0017327533302471335
     },
     {
       "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": 28742.242010498678,
-      "required_pages_per_second_upper_estimate": 0.002234582892554543
+      "required_token_positions_per_second": 28904.044439825793,
+      "required_pages_per_second_upper_estimate": 0.0022471623197410535
     },
     {
       "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": 28742.242010498678,
-      "required_pages_per_second_upper_estimate": 0.3686746194956283
+      "required_token_positions_per_second": 28904.044439825793,
+      "required_pages_per_second_upper_estimate": 0.3707500473291234
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0565046218828158,
-    "producer_p95_seconds": 0.1667780039997524,
-    "gpu_update_seconds": 0.961929135169797,
-    "producer_batches_per_second": 17.697667317797244,
-    "gpu_batches_per_second": 1.0395776190139858,
-    "producer_headroom": 17.02390181753148,
+    "producer_batch_seconds": 0.3247482479765509,
+    "producer_p95_seconds": 0.19076910900002986,
+    "gpu_update_seconds": 0.9565443361242851,
+    "producer_batches_per_second": 3.0793083757366633,
+    "gpu_batches_per_second": 1.045429848083977,
+    "producer_headroom": 2.9454949859909774,
     "sustainable": true,
-    "buffer_seconds": 60.60153551569721,
+    "buffer_seconds": 60.26229317582996,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 403707.9412238132,
-    "upload_bytes_per_second": 5034472.692620023,
+    "generation_bytes_per_second": 373381.23589013086,
+    "upload_bytes_per_second": 5138244.383784669,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.6263263467388,
+    "estimated_full_shards_per_hour": 80.1189213516993,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 6214716,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.56234888400013,
+      "bytes": 11134116,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 889000, "loss": 0.4985983073711395, "lr": 3e-05, "grad_norm": 0.312134335490036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130124948652, "tokens_per_second": 153313.03883898287, "data_wait_s": 2.075199972750852e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 889100, "loss": 0.5271111130714417, "lr": 3e-05, "grad_norm": 0.3237032503284707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130139706737, "tokens_per_second": 152814.1421765432, "data_wait_s": 1.918299949466018e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 889200, "loss": 0.5401551127433777, "lr": 3e-05, "grad_norm": 0.31936998721257415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130154458253, "tokens_per_second": 153469.32233748038, "data_wait_s": 9.684000360721257e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 889300, "loss": 0.5247880816459656, "lr": 3e-05, "grad_norm": 0.30831380885092613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130169050578, "tokens_per_second": 152004.8896396374, "data_wait_s": 1.3217999367043376e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3a040ec6959e239fe5a4fe3cb6ad25a43e2b66a10380fa09c608e3b05dbadb03
-size 89121
+oid sha256:9dbe8050a9583e910543152dbf38be6676f2b044b9e810d1bf9bddc26f1c8610
+size 91239