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": 74187.06921661421,
-      "required_pages_per_second_upper_estimate": 0.006430347134288202
+      "required_token_positions_per_second": 74523.22507601358,
+      "required_pages_per_second_upper_estimate": 0.006459484272201703
     },
     {
       "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": 9273.383652076776,
-      "required_pages_per_second_upper_estimate": 0.003699291388254658
+      "required_token_positions_per_second": 9315.403134501697,
+      "required_pages_per_second_upper_estimate": 0.003716053588041207
     },
     {
       "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": 18546.767304153553,
-      "required_pages_per_second_upper_estimate": 0.004290468203859724
+      "required_token_positions_per_second": 18630.806269003395,
+      "required_pages_per_second_upper_estimate": 0.004309909139342469
     },
     {
       "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": 27820.15095623033,
-      "required_pages_per_second_upper_estimate": 0.0016941012321828241
+      "required_token_positions_per_second": 27946.209403505094,
+      "required_pages_per_second_upper_estimate": 0.0017017775302442984
     },
     {
       "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": 27820.15095623033,
-      "required_pages_per_second_upper_estimate": 0.0021416722317173877
+      "required_token_positions_per_second": 27946.209403505094,
+      "required_pages_per_second_upper_estimate": 0.0021513765599407153
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27820.15095623033,
-      "required_pages_per_second_upper_estimate": 0.3454331668205959
+      "required_token_positions_per_second": 27946.209403505094,
+      "required_pages_per_second_upper_estimate": 0.34699839084526485
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4358273253208722,
-    "producer_p95_seconds": 0.4113335799993365,
-    "gpu_update_seconds": 0.9938120049563651,
-    "producer_batches_per_second": 2.294486696683745,
-    "gpu_batches_per_second": 1.006226524747914,
-    "producer_headroom": 2.28028842437698,
+    "producer_batch_seconds": 0.44000728884384444,
+    "producer_p95_seconds": 0.43699579200620065,
+    "gpu_update_seconds": 0.9893291644960017,
+    "producer_batches_per_second": 2.272689624364139,
+    "gpu_batches_per_second": 1.0107859303929794,
+    "producer_headroom": 2.2484381272309055,
     "sustainable": true,
-    "buffer_seconds": 60.62253230233827,
+    "buffer_seconds": 60.34907903425611,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305153.4206387143,
-    "upload_bytes_per_second": 6305064.137133642,
+    "generation_bytes_per_second": 289070.5973704777,
+    "upload_bytes_per_second": 6372235.538055288,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.47882046099731,
+    "estimated_full_shards_per_hour": 62.02782097659824,
     "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": 1.4500177529989742,
-      "bytes": 6186157,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1433413880004082,
       "bytes": 10345973,
@@ -562,6 +556,12 @@
       "bytes": 2405359,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.4177541469980497,
+      "bytes": 10495082,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -386,3 +386,4 @@
 {"step": 78600, "loss": 0.5955580398440361, "lr": 0.0003, "grad_norm": 0.14015123064081028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11502607001, "tokens_per_second": 147614.10810078675, "data_wait_s": 6.9829984568059444e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 78700, "loss": 0.6064422205090523, "lr": 0.0003, "grad_norm": 0.12661287513965544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11517252858, "tokens_per_second": 148433.22288894854, "data_wait_s": 7.493996236007661e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 78800, "loss": 0.6138870902359486, "lr": 0.0003, "grad_norm": 0.13637786705557958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11532109224, "tokens_per_second": 148779.2747232659, "data_wait_s": 1.252999936696142e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 78900, "loss": 0.6395603455603123, "lr": 0.0003, "grad_norm": 0.1199310508753388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11546839230, "tokens_per_second": 149588.12416878866, "data_wait_s": 8.640003215987235e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a132995d361bd7e747b9450dd0f9e68139cb348a1f337f89fe56de907a68bae3
-size 827430
+oid sha256:d1b437fa3e58e44e69222443e8aeeeb76ee7f9f32042d1b9245f882d0fc07998
+size 829548