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": 76292.80021700023,
-      "required_pages_per_second_upper_estimate": 0.0052781129426976835
+      "required_token_positions_per_second": 76048.02818102209,
+      "required_pages_per_second_upper_estimate": 0.0052611790452993445
     },
     {
       "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": 9536.600027125029,
-      "required_pages_per_second_upper_estimate": 0.003992775284502939
+      "required_token_positions_per_second": 9506.003522627761,
+      "required_pages_per_second_upper_estimate": 0.003979965166997603
     },
     {
       "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": 19073.200054250057,
-      "required_pages_per_second_upper_estimate": 0.003972444577189967
+      "required_token_positions_per_second": 19012.007045255523,
+      "required_pages_per_second_upper_estimate": 0.0039596996871845746
     },
     {
       "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": 28609.800081375084,
-      "required_pages_per_second_upper_estimate": 0.0017151138302431345
+      "required_token_positions_per_second": 28518.01056788328,
+      "required_pages_per_second_upper_estimate": 0.0017096111890637727
     },
     {
       "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": 28609.800081375084,
-      "required_pages_per_second_upper_estimate": 0.002224286115115664
+      "required_token_positions_per_second": 28518.01056788328,
+      "required_pages_per_second_upper_estimate": 0.0022171498841810776
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28609.800081375084,
-      "required_pages_per_second_upper_estimate": 0.36697579663389496
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28518.01056788328,
+      "required_pages_per_second_upper_estimate": 0.36579841931072304
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10407186764081189,
-    "producer_p95_seconds": 0.2944806070008781,
-    "gpu_update_seconds": 0.9663821460255078,
-    "producer_batches_per_second": 9.608744636459747,
-    "gpu_batches_per_second": 1.0347873293321428,
-    "producer_headroom": 9.285719262393059,
+    "producer_batch_seconds": 0.1017209329059483,
+    "producer_p95_seconds": 0.29111436200037133,
+    "gpu_update_seconds": 0.9694925925561203,
+    "producer_batches_per_second": 9.830818214424017,
+    "gpu_batches_per_second": 1.0314673961184637,
+    "producer_headroom": 9.53090543764987,
     "sustainable": true,
-    "buffer_seconds": 60.88207519960699,
+    "buffer_seconds": 60.10854073847946,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 481026.3838989734,
-    "upload_bytes_per_second": 5133788.348274914,
+    "generation_bytes_per_second": 491164.47118295816,
+    "upload_bytes_per_second": 5103377.552113862,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.21706426360036,
+    "estimated_full_shards_per_hour": 105.3924617921501,
     "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": 1.3326452839974081,
-      "bytes": 5962377,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.016969195999991,
       "bytes": 1171347,
@@ -562,6 +556,12 @@
       "bytes": 8373554,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.806914691995189,
+      "bytes": 5833325,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -336,3 +336,4 @@
 {"step": 753600, "loss": 0.4969802498817444, "lr": 5e-05, "grad_norm": 0.2267397935405334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110305946145, "tokens_per_second": 151113.381995992, "data_wait_s": 1.5074998373165727e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 753700, "loss": 0.5267372131347656, "lr": 5e-05, "grad_norm": 0.24205623519582248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110320734202, "tokens_per_second": 152452.91525136866, "data_wait_s": 1.2331001926213503e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 753800, "loss": 0.5201202034950256, "lr": 5e-05, "grad_norm": 0.24850761298632812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110335383790, "tokens_per_second": 151472.20133507627, "data_wait_s": 1.4154000382404774e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 753900, "loss": 0.53907310962677, "lr": 5e-05, "grad_norm": 0.24213909450658427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110349989328, "tokens_per_second": 150317.22320431276, "data_wait_s": 1.3217999367043376e-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:ec2cce82e6db909906233524cac1ac4ef6ce0e641e2b8668dd0e52bed45921ee
-size 720747
+oid sha256:c50e901b9010946a9afce458a81e86cbce4f1b7072977518a17029c479cfa35b
+size 722865