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": 77138.66069948499,
-      "required_pages_per_second_upper_estimate": 0.005336631533542701
+      "required_token_positions_per_second": 76410.81880324664,
+      "required_pages_per_second_upper_estimate": 0.005286277742335059
     },
     {
       "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": 9642.332587435623,
-      "required_pages_per_second_upper_estimate": 0.004037043299558052
+      "required_token_positions_per_second": 9551.35235040583,
+      "required_pages_per_second_upper_estimate": 0.00399895177419707
     },
     {
       "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": 19284.665174871247,
-      "required_pages_per_second_upper_estimate": 0.004016487185105113
+      "required_token_positions_per_second": 19102.70470081166,
+      "required_pages_per_second_upper_estimate": 0.003978589616978895
     },
     {
       "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": 28926.99776230687,
-      "required_pages_per_second_upper_estimate": 0.00173412934688218
+      "required_token_positions_per_second": 28654.05705121749,
+      "required_pages_per_second_upper_estimate": 0.0017177669680087065
     },
     {
       "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": 28926.99776230687,
-      "required_pages_per_second_upper_estimate": 0.0022489468396029617
+      "required_token_positions_per_second": 28654.05705121749,
+      "required_pages_per_second_upper_estimate": 0.0022277268998550735
     },
     {
       "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": 28926.99776230687,
-      "required_pages_per_second_upper_estimate": 0.3710444679045532
+      "required_token_positions_per_second": 28654.05705121749,
+      "required_pages_per_second_upper_estimate": 0.3675434775236014
     }
   ],
   "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": 0.1378222577501731,
-    "producer_p95_seconds": 0.31258224199700635,
-    "gpu_update_seconds": 0.9557853264684986,
-    "producer_batches_per_second": 7.255722089625571,
-    "gpu_batches_per_second": 1.0462600463797336,
-    "producer_headroom": 6.934912706197474,
+    "producer_batch_seconds": 0.11156623471126181,
+    "producer_p95_seconds": 0.29063131699876976,
+    "gpu_update_seconds": 0.9648895425377558,
+    "producer_batches_per_second": 8.963285375616044,
+    "gpu_batches_per_second": 1.0363880588547993,
+    "producer_headroom": 8.648580325713521,
     "sustainable": true,
-    "buffer_seconds": 60.21447556751541,
+    "buffer_seconds": 60.78804117987862,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351642.16822637,
-    "upload_bytes_per_second": 5181004.5318630515,
+    "generation_bytes_per_second": 366332.61916101706,
+    "upload_bytes_per_second": 5190132.985339433,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.45422349065137,
+    "estimated_full_shards_per_hour": 78.60645228503117,
     "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": 1.2533954940008698,
-      "bytes": 6219875,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2242291910006315,
       "bytes": 6745128,
@@ -562,6 +556,12 @@
       "bytes": 1341723,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.3343794770007662,
+      "bytes": 7400925,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -270,3 +270,4 @@
 {"step": 952000, "loss": 0.5298699736595154, "lr": 3e-05, "grad_norm": 0.36111805114874257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139346666069, "tokens_per_second": 150348.00721994363, "data_wait_s": 1.2268999853404239e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952100, "loss": 0.5685331225395203, "lr": 3e-05, "grad_norm": 0.5033133254657745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139361327151, "tokens_per_second": 152078.57846030113, "data_wait_s": 1.6996997146634385e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952200, "loss": 0.5294348001480103, "lr": 3e-05, "grad_norm": 0.3469594693218845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139376085306, "tokens_per_second": 153018.2428823329, "data_wait_s": 1.1668998922687024e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 952300, "loss": 0.5251018404960632, "lr": 3e-05, "grad_norm": 0.34399387024216427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139390817641, "tokens_per_second": 152716.52592594008, "data_wait_s": 1.2970998795935884e-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 sha256:2733bf6d9b43711ebcf258dd716c5bd1ca7d8cbc44d23e5f32aa2af25abef412
-size 580176
+oid sha256:91751b0bd9f583ec5337212dc1084efe94f2db1a32fec6f152de71b98df7019d
+size 582294