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": 77262.21989109206,
-      "required_pages_per_second_upper_estimate": 0.005345179645114902
+      "required_token_positions_per_second": 77030.20031474551,
+      "required_pages_per_second_upper_estimate": 0.005329127992463659
     },
     {
       "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": 9657.777486386507,
-      "required_pages_per_second_upper_estimate": 0.004043509756222622
+      "required_token_positions_per_second": 9628.775039343189,
+      "required_pages_per_second_upper_estimate": 0.004031367037285548
     },
     {
       "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": 19315.554972773014,
-      "required_pages_per_second_upper_estimate": 0.0040229207153893
+      "required_token_positions_per_second": 19257.550078686378,
+      "required_pages_per_second_upper_estimate": 0.004010839825642978
     },
     {
       "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": 28973.332459159523,
-      "required_pages_per_second_upper_estimate": 0.001736907041209511
+      "required_token_positions_per_second": 28886.325118029567,
+      "required_pages_per_second_upper_estimate": 0.0017316910839612876
     },
     {
       "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": 28973.332459159523,
-      "required_pages_per_second_upper_estimate": 0.002252549158478463
+      "required_token_positions_per_second": 28886.325118029567,
+      "required_pages_per_second_upper_estimate": 0.002245784720410457
     },
     {
       "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": 28973.332459159523,
-      "required_pages_per_second_upper_estimate": 0.37163879964545765
+      "required_token_positions_per_second": 28886.325118029567,
+      "required_pages_per_second_upper_estimate": 0.3705227628946469
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.2963794690076895,
-    "producer_p95_seconds": 0.279459200999554,
-    "gpu_update_seconds": 0.9542568166424178,
-    "producer_batches_per_second": 3.37405287669928,
-    "gpu_batches_per_second": 1.0479359251721472,
-    "producer_headroom": 3.219712957302247,
+    "producer_batch_seconds": 0.08033457020323453,
+    "producer_p95_seconds": 0.2824178099981509,
+    "gpu_update_seconds": 0.9571310953203715,
+    "producer_batches_per_second": 12.447941122609462,
+    "gpu_batches_per_second": 1.0447889582620649,
+    "producer_headroom": 11.91431152116669,
     "sustainable": true,
-    "buffer_seconds": 60.11817944847232,
+    "buffer_seconds": 60.29925900518341,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 433590.8678504284,
-    "upload_bytes_per_second": 5425746.375571762,
+    "generation_bytes_per_second": 441098.0428161206,
+    "upload_bytes_per_second": 5420217.19089,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.03850676188124,
+    "estimated_full_shards_per_hour": 94.64937175142968,
     "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": 0.8907613930023217,
-      "bytes": 6205142,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1056236590011395,
       "bytes": 6648792,
@@ -562,6 +556,12 @@
       "bytes": 8419016,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9963944320006703,
+      "bytes": 6337215,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -273,3 +273,4 @@
 {"step": 467500, "loss": 0.5648924708366394, "lr": 0.0003, "grad_norm": 0.18205196700194531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68428158331, "tokens_per_second": 152511.78075998984, "data_wait_s": 9.354000212624669e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 467600, "loss": 0.5514411926269531, "lr": 0.0003, "grad_norm": 0.15295055775286695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68442647133, "tokens_per_second": 150059.95207659356, "data_wait_s": 1.016999885905534e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 467700, "loss": 0.592983067035675, "lr": 0.0003, "grad_norm": 0.15200497079987607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68457398941, "tokens_per_second": 153420.24164965306, "data_wait_s": 9.367002348881215e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 467800, "loss": 0.5810258388519287, "lr": 0.0003, "grad_norm": 0.17461217265829648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68472102331, "tokens_per_second": 152561.36892940756, "data_wait_s": 1.8931001250166446e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fbc7c8a74ba12f6aab190bc509488c0a78b6f0445425ed741e5b5783fdf6869
-size 586530
+oid sha256:0e43b8edb43c5d423d7f6674f1452b12d75c93f138274bf8ecf724c3df4b39b4
+size 588648