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": 76742.60649064253,
-      "required_pages_per_second_upper_estimate": 0.0053092315844026055
+      "required_token_positions_per_second": 76497.7673544901,
+      "required_pages_per_second_upper_estimate": 0.005292293044858515
     },
     {
       "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": 9592.825811330316,
-      "required_pages_per_second_upper_estimate": 0.004016315846221805
+      "required_token_positions_per_second": 9562.220919311263,
+      "required_pages_per_second_upper_estimate": 0.004003502217036247
     },
     {
       "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": 19185.651622660633,
-      "required_pages_per_second_upper_estimate": 0.003995865273342607
+      "required_token_positions_per_second": 19124.441838622526,
+      "required_pages_per_second_upper_estimate": 0.003983116889538026
     },
     {
       "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": 28778.47743399095,
-      "required_pages_per_second_upper_estimate": 0.001725225779977051
+      "required_token_positions_per_second": 28686.662757933787,
+      "required_pages_per_second_upper_estimate": 0.001719721630340317
     },
     {
       "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": 28778.47743399095,
-      "required_pages_per_second_upper_estimate": 0.0022374000373482834
+      "required_token_positions_per_second": 28686.662757933787,
+      "required_pages_per_second_upper_estimate": 0.002230261850134919
     },
     {
       "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": 28778.47743399095,
-      "required_pages_per_second_upper_estimate": 0.36913940860162064
+      "required_token_positions_per_second": 28686.662757933787,
+      "required_pages_per_second_upper_estimate": 0.3679617085200778
     }
   ],
   "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": 0.13259479776556304,
-    "producer_p95_seconds": 0.27210508300049696,
-    "gpu_update_seconds": 0.9607179554031682,
-    "producer_batches_per_second": 7.541774012643169,
-    "gpu_batches_per_second": 1.0408882173752514,
-    "producer_headroom": 7.2455177095392935,
+    "producer_batch_seconds": 0.07635977181254816,
+    "producer_p95_seconds": 0.2760391219999292,
+    "gpu_update_seconds": 0.9637928340881503,
+    "producer_batches_per_second": 13.095900842329003,
+    "gpu_batches_per_second": 1.0375673740572118,
+    "producer_headroom": 12.621735387765666,
     "sustainable": true,
-    "buffer_seconds": 60.525231190399595,
+    "buffer_seconds": 60.71894854755347,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269166.84837642295,
-    "upload_bytes_per_second": 5291585.741498411,
+    "generation_bytes_per_second": 269564.9901411563,
+    "upload_bytes_per_second": 5249635.761448413,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.75693977803723,
+    "estimated_full_shards_per_hour": 57.84237173248307,
     "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": 1.0690768630011007,
-      "bytes": 5982757,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3195378630007326,
       "bytes": 8379086,
@@ -562,6 +556,12 @@
       "bytes": 6031296,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0317238650022773,
+      "bytes": 2334152,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -286,3 +286,4 @@
 {"step": 748600, "loss": 0.514732837677002, "lr": 5e-05, "grad_norm": 0.24052960704157414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109574050392, "tokens_per_second": 150135.04202254055, "data_wait_s": 8.305996743729338e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 748700, "loss": 0.5358645915985107, "lr": 5e-05, "grad_norm": 0.23892753755651322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109588654043, "tokens_per_second": 150922.2132229246, "data_wait_s": 1.5794998034834862e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 748800, "loss": 0.5097352862358093, "lr": 5e-05, "grad_norm": 0.2574977276468063, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109603179872, "tokens_per_second": 149288.6030433589, "data_wait_s": 1.4174001989886165e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 748900, "loss": 0.5106076598167419, "lr": 5e-05, "grad_norm": 0.23015109633140107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109617845958, "tokens_per_second": 151143.99835627805, "data_wait_s": 1.2545002391561866e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ae3d36cf767a6319a9f4f36e5f2b42a4da49d01df0f9d1f7c4f9a3cecd16bbc
-size 614064
+oid sha256:3cd6eb45561f5039f89a0a2c979976609d313542ccaed1f50cfb55ea26086639
+size 616182