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": 77183.25214935331,
-      "required_pages_per_second_upper_estimate": 0.00533971647351101
+      "required_token_positions_per_second": 76706.60857997264,
+      "required_pages_per_second_upper_estimate": 0.005306741165415803
     },
     {
       "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": 9647.906518669164,
-      "required_pages_per_second_upper_estimate": 0.004039376988168615
+      "required_token_positions_per_second": 9588.32607249658,
+      "required_pages_per_second_upper_estimate": 0.0040144318995373515
     },
     {
       "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": 19295.81303733833,
-      "required_pages_per_second_upper_estimate": 0.004018808990868103
+      "required_token_positions_per_second": 19176.65214499316,
+      "required_pages_per_second_upper_estimate": 0.003993990919476658
     },
     {
       "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": 28943.719556007494,
-      "required_pages_per_second_upper_estimate": 0.0017351317928818318
+      "required_token_positions_per_second": 28764.97821748974,
+      "required_pages_per_second_upper_estimate": 0.0017244165225599138
     },
     {
       "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": 28943.719556007494,
-      "required_pages_per_second_upper_estimate": 0.002250246885511788
+      "required_token_positions_per_second": 28764.97821748974,
+      "required_pages_per_second_upper_estimate": 0.0022363505326420926
     },
     {
       "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": 28943.719556007494,
-      "required_pages_per_second_upper_estimate": 0.37125895711968154
+      "required_token_positions_per_second": 28764.97821748974,
+      "required_pages_per_second_upper_estimate": 0.3689662551466726
     }
   ],
   "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": 0.3614161188126275,
-    "producer_p95_seconds": 0.18403391699939675,
-    "gpu_update_seconds": 0.9552331360349103,
-    "producer_batches_per_second": 2.7668937491922985,
-    "gpu_batches_per_second": 1.0468648566264285,
-    "producer_headroom": 2.64302859311635,
+    "producer_batch_seconds": 0.07215003269543274,
+    "producer_p95_seconds": 0.19292236299952492,
+    "gpu_update_seconds": 0.9611688140681228,
+    "producer_batches_per_second": 13.860007579224593,
+    "gpu_batches_per_second": 1.0403999644636046,
+    "producer_headroom": 13.321807047898496,
     "sustainable": true,
-    "buffer_seconds": 60.17968757019935,
+    "buffer_seconds": 60.55363528629174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319899.87026104267,
-    "upload_bytes_per_second": 5370585.742562781,
+    "generation_bytes_per_second": 292051.9646232365,
+    "upload_bytes_per_second": 5552366.120279601,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.64306527017078,
+    "estimated_full_shards_per_hour": 62.66755298636266,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 7199351,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.3326483809996716,
+      "bytes": 16444186,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -85,3 +85,4 @@
 {"step": 768500, "loss": 0.5103076100349426, "lr": 3e-05, "grad_norm": 0.25224451025330713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112485802586, "tokens_per_second": 151647.1536539731, "data_wait_s": 1.4532000932376832e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 768600, "loss": 0.5482544302940369, "lr": 3e-05, "grad_norm": 0.24336627572951555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112500311605, "tokens_per_second": 151309.71868873778, "data_wait_s": 1.6484000298078172e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 768700, "loss": 0.48764196038246155, "lr": 3e-05, "grad_norm": 0.24779002240778725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112514857299, "tokens_per_second": 150597.52495925015, "data_wait_s": 1.3955001122667454e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 768800, "loss": 0.5474236011505127, "lr": 3e-05, "grad_norm": 0.2541554534120326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112529532882, "tokens_per_second": 152436.10863425204, "data_wait_s": 1.85289991350146e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e23470157babe4520bec206076025e6635308d887b445b0518f4d7e874410a2f
-size 185214
+oid sha256:06b273e5ee775b88d624accc88ed65fd3e7d299a5e45d0778e85b8bc1299ffff
+size 187332