Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75641.37147579817,
-      "required_pages_per_second_upper_estimate": 0.00655640236821889
+      "required_token_positions_per_second": 75821.16511049683,
+      "required_pages_per_second_upper_estimate": 0.006571986424791769
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9455.17143447477,
-      "required_pages_per_second_upper_estimate": 0.0037718092526227744
+      "required_token_positions_per_second": 9477.645638812104,
+      "required_pages_per_second_upper_estimate": 0.0037807745487522357
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18910.34286894954,
-      "required_pages_per_second_upper_estimate": 0.004374575012063879
+      "required_token_positions_per_second": 18955.29127762421,
+      "required_pages_per_second_upper_estimate": 0.004384973035345788
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28365.514303424312,
-      "required_pages_per_second_upper_estimate": 0.0017273109987265879
+      "required_token_positions_per_second": 28432.936916436312,
+      "required_pages_per_second_upper_estimate": 0.0017314166821199067
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28365.514303424312,
-      "required_pages_per_second_upper_estimate": 0.0021836558118467483
+      "required_token_positions_per_second": 28432.936916436312,
+      "required_pages_per_second_upper_estimate": 0.0021888461912341504
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28365.514303424312,
-      "required_pages_per_second_upper_estimate": 0.35220475437903465
+      "required_token_positions_per_second": 28432.936916436312,
+      "required_pages_per_second_upper_estimate": 0.3530419175836735
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.08002749911716478,
-    "producer_p95_seconds": 0.27163532600025064,
-    "gpu_update_seconds": 0.974704696140916,
-    "producer_batches_per_second": 12.495704739391437,
-    "gpu_batches_per_second": 1.025951761553252,
-    "producer_headroom": 12.179622091075135,
+    "producer_batch_seconds": 0.10636465545310614,
+    "producer_p95_seconds": 0.2677414930003579,
+    "gpu_update_seconds": 0.9723933929603115,
+    "producer_batches_per_second": 9.401619323073708,
+    "gpu_batches_per_second": 1.028390368794716,
+    "producer_headroom": 9.142072512884871,
     "sustainable": true,
-    "buffer_seconds": 60.4316911607368,
+    "buffer_seconds": 60.288390363539314,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422234.36233811744,
-    "upload_bytes_per_second": 5779611.968227663,
+    "generation_bytes_per_second": 407991.0094780362,
+    "upload_bytes_per_second": 5753543.420760775,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.60166504485743,
+    "estimated_full_shards_per_hour": 87.54537308936895,
     "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": 1.107625275000828,
-      "bytes": 7585534,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.0756502530002763,
       "bytes": 6680691,
@@ -562,6 +556,12 @@
       "bytes": 6934101,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.4433788420028577,
+      "bytes": 7611195,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -158,3 +158,4 @@
 {"step": 370800, "loss": 0.54385956376791, "lr": 0.0003, "grad_norm": 0.18395404622296668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54274500728, "tokens_per_second": 149892.2020614638, "data_wait_s": 9.662000593380071e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 370900, "loss": 0.6035262010991573, "lr": 0.0003, "grad_norm": 0.16489540161898184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54289111778, "tokens_per_second": 149118.69607542423, "data_wait_s": 1.31840006361017e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 371000, "loss": 0.5991155337542295, "lr": 0.0003, "grad_norm": 0.1632551389304045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54303601846, "tokens_per_second": 148165.97946345946, "data_wait_s": 1.1613003152888268e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 371100, "loss": 0.5825608931481838, "lr": 0.0003, "grad_norm": 0.13730084919229854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54318136898, "tokens_per_second": 147243.17809376516, "data_wait_s": 1.3117001799400896e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1066f049c90206eafdb596825fb562592a1cd9bf536a8dd2f922727b46e3633c
-size 341394
+oid sha256:e215f87f229b8ffbb30e803237641a1ac4a7b0024dd93c0a258086efb5a0cb2d
+size 343512