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": 75858.13281100985,
-      "required_pages_per_second_upper_estimate": 0.006575190691378457
+      "required_token_positions_per_second": 76092.91144212967,
+      "required_pages_per_second_upper_estimate": 0.006595540708082923
     },
     {
       "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": 9482.266601376232,
-      "required_pages_per_second_upper_estimate": 0.0037826179198086132
+      "required_token_positions_per_second": 9511.613930266209,
+      "required_pages_per_second_upper_estimate": 0.0037943250080844934
     },
     {
       "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": 18964.533202752464,
-      "required_pages_per_second_upper_estimate": 0.0043871109920718835
+      "required_token_positions_per_second": 19023.227860532417,
+      "required_pages_per_second_upper_estimate": 0.004400688968158579
     },
     {
       "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": 28446.799804128696,
-      "required_pages_per_second_upper_estimate": 0.0017322608592474221
+      "required_token_positions_per_second": 28534.84179079863,
+      "required_pages_per_second_upper_estimate": 0.0017376221543150161
     },
     {
       "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": 28446.799804128696,
-      "required_pages_per_second_upper_estimate": 0.002189913394703639
+      "required_token_positions_per_second": 28534.84179079863,
+      "required_pages_per_second_upper_estimate": 0.002196691110553308
     },
     {
       "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": 28446.799804128696,
-      "required_pages_per_second_upper_estimate": 0.35321404825271235
+      "required_token_positions_per_second": 28534.84179079863,
+      "required_pages_per_second_upper_estimate": 0.35430723506957834
     }
   ],
   "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": 0.5191029269453225,
-    "producer_p95_seconds": 0.3569480099977227,
-    "gpu_update_seconds": 0.971919519607518,
-    "producer_batches_per_second": 1.926400234120296,
-    "gpu_batches_per_second": 1.0288917753229418,
-    "producer_headroom": 1.8723059901180081,
+    "producer_batch_seconds": 0.09469398521090966,
+    "producer_p95_seconds": 0.28461746299944934,
+    "gpu_update_seconds": 0.9689207391686118,
+    "producer_batches_per_second": 10.560332821273958,
+    "gpu_batches_per_second": 1.0320761643083995,
+    "producer_headroom": 10.232125483055315,
     "sustainable": true,
-    "buffer_seconds": 60.259010215666116,
+    "buffer_seconds": 60.07308582845393,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374753.9204028042,
+    "generation_bytes_per_second": 388859.7440256452,
     "upload_bytes_per_second": 5572694.160781924,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.41346749365897,
+    "estimated_full_shards_per_hour": 83.44024887635247,
     "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

@@ -221,3 +221,4 @@
 {"step": 177100, "loss": 0.6259839870035648, "lr": 0.0003, "grad_norm": 0.14912587300424573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25919474028, "tokens_per_second": 150611.38927202567, "data_wait_s": 1.5474997780984268e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 177200, "loss": 0.5881300400942564, "lr": 0.0003, "grad_norm": 0.1371592981511136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25934216054, "tokens_per_second": 138680.48862020616, "data_wait_s": 1.2162003258708864e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 177300, "loss": 0.5884219035506248, "lr": 0.0003, "grad_norm": 0.13482742427755728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25948848126, "tokens_per_second": 149604.35016829346, "data_wait_s": 1.1733998690033332e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 177400, "loss": 0.5940247401595116, "lr": 0.0003, "grad_norm": 0.13568014934017145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25963575225, "tokens_per_second": 149906.75099345375, "data_wait_s": 1.5060002624522895e-05, "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:e81205539b36f430c4a059161393e853c8a857a2d8183cd1df45ea639a9ed98e
-size 475611
+oid sha256:6cce777751380cf4a7fa858b4e150a8f9a92f3a5508746136f89d2fc6f6121c2
+size 477729