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": 77101.74535515234,
-      "required_pages_per_second_upper_estimate": 0.005334077644366369
+      "required_token_positions_per_second": 76735.35065220822,
+      "required_pages_per_second_upper_estimate": 0.005308729608663855
     },
     {
       "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": 9637.718169394042,
-      "required_pages_per_second_upper_estimate": 0.004035111339084048
+      "required_token_positions_per_second": 9591.918831526027,
+      "required_pages_per_second_upper_estimate": 0.004015936112717641
     },
     {
       "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": 19275.436338788084,
-      "required_pages_per_second_upper_estimate": 0.004014565061929765
+      "required_token_positions_per_second": 19183.837663052054,
+      "required_pages_per_second_upper_estimate": 0.003995487473393472
     },
     {
       "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": 28913.154508182124,
-      "required_pages_per_second_upper_estimate": 0.0017332994649348267
+      "required_token_positions_per_second": 28775.756494578083,
+      "required_pages_per_second_upper_estimate": 0.0017250626638138837
     },
     {
       "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": 28913.154508182124,
-      "required_pages_per_second_upper_estimate": 0.0022478705874848025
+      "required_token_positions_per_second": 28775.756494578083,
+      "required_pages_per_second_upper_estimate": 0.0022371884962770804
     },
     {
       "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": 28913.154508182124,
-      "required_pages_per_second_upper_estimate": 0.3708669015043692
+      "required_token_positions_per_second": 28775.756494578083,
+      "required_pages_per_second_upper_estimate": 0.36910450731234956
     }
   ],
   "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": 0.23039847503923738,
-    "producer_p95_seconds": 0.3205673660049797,
-    "gpu_update_seconds": 0.9562429444416347,
-    "producer_batches_per_second": 4.340306505195826,
-    "gpu_batches_per_second": 1.045759349977652,
-    "producer_headroom": 4.150387472307638,
+    "producer_batch_seconds": 0.1069208343592436,
+    "producer_p95_seconds": 0.34090551199915353,
+    "gpu_update_seconds": 0.9608087976838915,
+    "producer_batches_per_second": 9.352714145870742,
+    "gpu_batches_per_second": 1.0407898037680152,
+    "producer_headroom": 8.98617003357519,
     "sustainable": true,
-    "buffer_seconds": 60.24330549982299,
+    "buffer_seconds": 60.53095425408517,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493123.84616744716,
-    "upload_bytes_per_second": 4828279.47152803,
+    "generation_bytes_per_second": 495105.83389838005,
+    "upload_bytes_per_second": 4820500.851571604,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.81289805190622,
+    "estimated_full_shards_per_hour": 106.2381864806514,
     "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": 1.577455007001845,
-      "bytes": 11622766,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.0520612039981643,
       "bytes": 811209,
@@ -562,6 +556,12 @@
       "bytes": 2451644,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.337224768998567,
+      "bytes": 9767269,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -336,3 +336,4 @@
 {"step": 958600, "loss": 0.5070256590843201, "lr": 3e-05, "grad_norm": 0.3345787106480922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140313293116, "tokens_per_second": 152514.82561405018, "data_wait_s": 8.761999197304249e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 958700, "loss": 0.506074845790863, "lr": 3e-05, "grad_norm": 0.3511785905832599, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140327799300, "tokens_per_second": 151487.13559697891, "data_wait_s": 1.4682002074550837e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 958800, "loss": 0.5966412425041199, "lr": 3e-05, "grad_norm": 0.34871033800363765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140342284804, "tokens_per_second": 150215.1614174971, "data_wait_s": 1.3156000932212919e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 958900, "loss": 0.502213716506958, "lr": 3e-05, "grad_norm": 0.3423945327166306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140356824028, "tokens_per_second": 151072.07346903824, "data_wait_s": 1.2782998965121806e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:56f10a0dc98fb0cb9c4c692954c9e165dccc23172d9b6b69fe6359ec25d9e0c0
-size 720747
+oid sha256:7b5133701cc8abd1d3bf51c2209ec042c13163df10254e5695a10cddb629e72f
+size 722865