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": 75831.22607772931,
-      "required_pages_per_second_upper_estimate": 0.006572858483932192
+      "required_token_positions_per_second": 75821.86282504725,
+      "required_pages_per_second_upper_estimate": 0.006572046900920658
     },
     {
       "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": 9478.903259716164,
-      "required_pages_per_second_upper_estimate": 0.003781276232533973
+      "required_token_positions_per_second": 9477.732853130907,
+      "required_pages_per_second_upper_estimate": 0.003780809339847976
     },
     {
       "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": 18957.806519432328,
-      "required_pages_per_second_upper_estimate": 0.004385554892271353
+      "required_token_positions_per_second": 18955.465706261813,
+      "required_pages_per_second_upper_estimate": 0.0043850133863412594
     },
     {
       "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": 28436.70977914849,
-      "required_pages_per_second_upper_estimate": 0.0017316464296643973
+      "required_token_positions_per_second": 28433.198559392724,
+      "required_pages_per_second_upper_estimate": 0.0017314326148032176
     },
     {
       "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": 28436.70977914849,
-      "required_pages_per_second_upper_estimate": 0.002189136636649686
+      "required_token_positions_per_second": 28433.198559392724,
+      "required_pages_per_second_upper_estimate": 0.0021888663332332086
     },
     {
       "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": 28436.70977914849,
-      "required_pages_per_second_upper_estimate": 0.3530887639115995
+      "required_token_positions_per_second": 28433.198559392724,
+      "required_pages_per_second_upper_estimate": 0.3530451663135295
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10326935037500107,
-    "producer_p95_seconds": 0.24360115099989343,
-    "gpu_update_seconds": 0.9722643799063273,
-    "producer_batches_per_second": 9.683415227932672,
-    "gpu_batches_per_second": 1.028526829396285,
-    "producer_headroom": 9.414839701961448,
+    "producer_batch_seconds": 0.0643923006797138,
+    "producer_p95_seconds": 0.21713291599735385,
+    "gpu_update_seconds": 0.9723844449736262,
+    "producer_batches_per_second": 15.529806971395274,
+    "gpu_batches_per_second": 1.0283998321539614,
+    "producer_headroom": 15.100942732427745,
     "sustainable": true,
-    "buffer_seconds": 60.28039155419229,
+    "buffer_seconds": 60.28783558836482,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443503.89601263136,
-    "upload_bytes_per_second": 5841700.966734531,
+    "generation_bytes_per_second": 444075.09937671485,
+    "upload_bytes_per_second": 5865446.35776112,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.16561184200484,
+    "estimated_full_shards_per_hour": 95.28817878700336,
     "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": [
@@ -442,6 +442,12 @@
       "bytes": 581892,
       "pages": 17,
       "shards": 1
+    },
+    {
+      "seconds": 1.1391355480009224,
+      "bytes": 7817543,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -100,3 +100,4 @@
 {"step": 95000, "loss": 0.6396308094263077, "lr": 0.0003, "grad_norm": 0.143320235435905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13902633353, "tokens_per_second": 148697.1748974744, "data_wait_s": 1.7082998965634033e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 95100, "loss": 0.6108560711145401, "lr": 0.0003, "grad_norm": 0.13584040699080668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13917151076, "tokens_per_second": 131626.97595572475, "data_wait_s": 1.3256998499855399e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 95200, "loss": 0.5910352412611246, "lr": 0.0003, "grad_norm": 0.12485652502635318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13931717642, "tokens_per_second": 148402.46917393446, "data_wait_s": 9.585000952938572e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 95300, "loss": 0.625568326562643, "lr": 0.0003, "grad_norm": 0.13915032065916308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13946370067, "tokens_per_second": 148887.99585646426, "data_wait_s": 1.2813001376343891e-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 sha256:441bd5df66fc4de6e5fe84a3f09c0fbd75f0bd5dab8c446d8c3cecfd56479324
-size 217767
+oid sha256:5dd398d863d92c51dbf561c3e6831eb9bed3a04d0a714ed4730a409d4138f0ff
+size 219885