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": 75048.65254625826,
-      "required_pages_per_second_upper_estimate": 0.006505026993638764
+      "required_token_positions_per_second": 76095.65271351882,
+      "required_pages_per_second_upper_estimate": 0.006595778314539245
     },
     {
       "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": 9381.081568282283,
-      "required_pages_per_second_upper_estimate": 0.003742253697256376
+      "required_token_positions_per_second": 9511.956589189853,
+      "required_pages_per_second_upper_estimate": 0.0037944616998523426
     },
     {
       "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": 18762.163136564566,
-      "required_pages_per_second_upper_estimate": 0.0043402962388508345
+      "required_token_positions_per_second": 19023.913178379706,
+      "required_pages_per_second_upper_estimate": 0.004400847504381368
     },
     {
       "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": 28143.24470484685,
-      "required_pages_per_second_upper_estimate": 0.0017137759463316787
+      "required_token_positions_per_second": 28535.86976756956,
+      "required_pages_per_second_upper_estimate": 0.0017376847527069885
     },
     {
       "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": 28143.24470484685,
-      "required_pages_per_second_upper_estimate": 0.0021665448828666264
+      "required_token_positions_per_second": 28535.86976756956,
+      "required_pages_per_second_upper_estimate": 0.002196770247050756
     },
     {
       "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": 28143.24470484685,
-      "required_pages_per_second_upper_estimate": 0.34944490985319604
+      "required_token_positions_per_second": 28535.86976756956,
+      "required_pages_per_second_upper_estimate": 0.3543199991006563
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12767865269552203,
-    "producer_p95_seconds": 0.40239784799632616,
-    "gpu_update_seconds": 0.9824027147530164,
-    "producer_batches_per_second": 7.832162847024405,
-    "gpu_batches_per_second": 1.0179124965584074,
-    "producer_headroom": 7.694338043304489,
+    "producer_batch_seconds": 0.25487266636775985,
+    "producer_p95_seconds": 0.4230957529944135,
+    "gpu_update_seconds": 0.9688858347475848,
+    "producer_batches_per_second": 3.923527831568584,
+    "gpu_batches_per_second": 1.0321133451811906,
+    "producer_headroom": 3.8014505382447092,
     "sustainable": true,
-    "buffer_seconds": 60.908968314687016,
+    "buffer_seconds": 60.07092175435026,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435647.91254290193,
+    "generation_bytes_per_second": 424775.5387203291,
     "upload_bytes_per_second": 4631406.143694578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.47990066733641,
+    "estimated_full_shards_per_hour": 91.14694234092146,
     "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": [

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -424,3 +424,4 @@
 {"step": 197400, "loss": 0.6113288514316082, "lr": 0.0003, "grad_norm": 0.14848443454253163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28891607788, "tokens_per_second": 149249.5310237459, "data_wait_s": 1.13159985630773e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 197500, "loss": 0.5889865458011627, "lr": 0.0003, "grad_norm": 0.13283690496451514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28906201905, "tokens_per_second": 148853.07201273882, "data_wait_s": 1.367599907098338e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 197600, "loss": 0.586609398946166, "lr": 0.0003, "grad_norm": 0.17234701445874742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28920839583, "tokens_per_second": 149601.59167833344, "data_wait_s": 1.1750998964998871e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 197700, "loss": 0.6029997207224369, "lr": 0.0003, "grad_norm": 0.15610074409238134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28935300702, "tokens_per_second": 147006.29566422856, "data_wait_s": 8.101000275928527e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6159c16b767b617946ee152b37a31a635a61f0cf29301914d2ad14b101dcec1
-size 908697
+oid sha256:0fb95d19ea3e3f4262d6931b81ae4571fca4c37457bdefe564c23cda0863b2c6
+size 910815