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": 76865.9105651411,
-      "required_pages_per_second_upper_estimate": 0.005317762046381287
+      "required_token_positions_per_second": 77020.92941428565,
+      "required_pages_per_second_upper_estimate": 0.005328486610058394
     },
     {
       "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": 9608.238820642637,
-      "required_pages_per_second_upper_estimate": 0.004022768951360639
+      "required_token_positions_per_second": 9627.616176785707,
+      "required_pages_per_second_upper_estimate": 0.004030881845732532
     },
     {
       "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": 19216.477641285273,
-      "required_pages_per_second_upper_estimate": 0.004002285520085341
+      "required_token_positions_per_second": 19255.232353571413,
+      "required_pages_per_second_upper_estimate": 0.0040103571046240525
     },
     {
       "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": 28824.71646192791,
-      "required_pages_per_second_upper_estimate": 0.0017279977390989643
+      "required_token_positions_per_second": 28882.848530357118,
+      "required_pages_per_second_upper_estimate": 0.0017314826678387664
     },
     {
       "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": 28824.71646192791,
-      "required_pages_per_second_upper_estimate": 0.0022409949183863926
+      "required_token_positions_per_second": 28882.848530357118,
+      "required_pages_per_second_upper_estimate": 0.002245514431010817
     },
     {
       "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": 28824.71646192791,
-      "required_pages_per_second_upper_estimate": 0.36973251320439593
+      "required_token_positions_per_second": 28882.848530357118,
+      "required_pages_per_second_upper_estimate": 0.3704781689608537
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.45215287286697503,
-    "producer_p95_seconds": 0.28928426199854584,
-    "gpu_update_seconds": 0.9591768243936716,
-    "producer_batches_per_second": 2.2116413717760532,
-    "gpu_batches_per_second": 1.042560635920425,
-    "producer_headroom": 2.1213551476778183,
+    "producer_batch_seconds": 0.10341664211708235,
+    "producer_p95_seconds": 0.29676626099899295,
+    "gpu_update_seconds": 0.9572463038380983,
+    "producer_batches_per_second": 9.66962356859216,
+    "gpu_batches_per_second": 1.0446632136269212,
+    "producer_headroom": 9.256211420540605,
     "sustainable": true,
-    "buffer_seconds": 60.42813993680131,
+    "buffer_seconds": 60.30651714180019,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349035.12211858423,
-    "upload_bytes_per_second": 4858761.642571554,
+    "generation_bytes_per_second": 335368.31758853584,
+    "upload_bytes_per_second": 4874240.16890082,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.89481208484789,
+    "estimated_full_shards_per_hour": 71.96223397962623,
     "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": 1.2773266180010978,
-      "bytes": 7212495,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.216828146003536,
       "bytes": 6096262,
@@ -562,6 +556,12 @@
       "bytes": 2264477,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.6527253510030278,
+      "bytes": 10383315,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -292,3 +292,4 @@
 {"step": 789200, "loss": 0.5413856506347656, "lr": 3e-05, "grad_norm": 0.272991260775642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115516445817, "tokens_per_second": 152943.70408638308, "data_wait_s": 1.0864998330362141e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 789300, "loss": 0.5332176089286804, "lr": 3e-05, "grad_norm": 0.26102109338624185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115530997889, "tokens_per_second": 151776.20846422215, "data_wait_s": 1.37959978019353e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 789400, "loss": 0.49407896399497986, "lr": 3e-05, "grad_norm": 0.2954638379966947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115545605874, "tokens_per_second": 152207.10726596077, "data_wait_s": 1.516799966339022e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 789500, "loss": 0.5112563371658325, "lr": 3e-05, "grad_norm": 0.2565419364300872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115560164229, "tokens_per_second": 150897.98640654597, "data_wait_s": 1.8455000827088952e-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 sha256:8390e601790ed8fad07cbdb86d7aba6dc3413c0214a15df700ffa08262070a32
-size 626772
+oid sha256:2f5212bfd5065824054610818bd62fdf36e6edf187c676e2f874f96ced014a3c
+size 628890