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": 76992.55578518612,
-      "required_pages_per_second_upper_estimate": 0.005326523656561398
+      "required_token_positions_per_second": 76627.61088291103,
+      "required_pages_per_second_upper_estimate": 0.00530127592143317
     },
     {
       "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": 9624.069473148265,
-      "required_pages_per_second_upper_estimate": 0.004029396914983129
+      "required_token_positions_per_second": 9578.451360363879,
+      "required_pages_per_second_upper_estimate": 0.0040102975637748275
     },
     {
       "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": 19248.13894629653,
-      "required_pages_per_second_upper_estimate": 0.004008879734954424
+      "required_token_positions_per_second": 19156.902720727758,
+      "required_pages_per_second_upper_estimate": 0.003989877635229517
     },
     {
       "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": 28872.2084194448,
-      "required_pages_per_second_upper_estimate": 0.0017308448094360816
+      "required_token_positions_per_second": 28735.354081091635,
+      "required_pages_per_second_upper_estimate": 0.0017226406008162844
     },
     {
       "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": 28872.2084194448,
-      "required_pages_per_second_upper_estimate": 0.002244687209188286
+      "required_token_positions_per_second": 28735.354081091635,
+      "required_pages_per_second_upper_estimate": 0.002234047386339945
     },
     {
       "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": 28872.2084194448,
-      "required_pages_per_second_upper_estimate": 0.3703416890425315
+      "required_token_positions_per_second": 28735.354081091635,
+      "required_pages_per_second_upper_estimate": 0.3685862685328771
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05058798225782368,
-    "producer_p95_seconds": 0.16410533500038582,
-    "gpu_update_seconds": 0.9575990723792254,
-    "producer_batches_per_second": 19.767540735336308,
-    "gpu_batches_per_second": 1.0442783716523727,
-    "producer_headroom": 18.929378671376597,
+    "producer_batch_seconds": 0.08861975876563122,
+    "producer_p95_seconds": 0.1585833499993896,
+    "gpu_update_seconds": 0.9621597117605336,
+    "producer_batches_per_second": 11.284165223747179,
+    "gpu_batches_per_second": 1.0393284896228165,
+    "producer_headroom": 10.857169159138822,
     "sustainable": true,
-    "buffer_seconds": 60.3287415598912,
+    "buffer_seconds": 60.616061840913616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 166,
+    "upload_pages_per_commit": 173,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 622966.3911863102,
-    "upload_bytes_per_second": 4247089.061955517,
+    "generation_bytes_per_second": 714567.3819208612,
+    "upload_bytes_per_second": 2955143.578231099,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.8370428166007,
+    "estimated_full_shards_per_hour": 76.66476294145286,
     "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": [
@@ -196,6 +196,12 @@
       "bytes": 5982767,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.7043199660001846,
+      "bytes": 963598,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 689400, "loss": 0.542589008808136, "lr": 8e-05, "grad_norm": 0.19721698587079897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100909441741, "tokens_per_second": 151855.8219929631, "data_wait_s": 1.6449999748147093e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689500, "loss": 0.5766230225563049, "lr": 8e-05, "grad_norm": 0.2064560455752057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100924019971, "tokens_per_second": 151795.06198272988, "data_wait_s": 1.6125000001920853e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689600, "loss": 0.5401948690414429, "lr": 8e-05, "grad_norm": 0.2001230868282645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100938656284, "tokens_per_second": 152377.85305167912, "data_wait_s": 1.9877999875461683e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 689700, "loss": 0.5680307745933533, "lr": 8e-05, "grad_norm": 0.19993649434375324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100953492215, "tokens_per_second": 154183.4022057356, "data_wait_s": 1.3666999620909337e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dcdb6665b89e107c51b83317f04de5a8ceb18f8daac3f6cfa3a2e1d73cf56f0b
-size 97593
+oid sha256:5135728728e0eeb4dd7e2793e10ac084773bbe219ab13914c8ab352601aeb786
+size 99711