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": 76356.4831590772,
-      "required_pages_per_second_upper_estimate": 0.005282518676395352
+      "required_token_positions_per_second": 76706.39890162514,
+      "required_pages_per_second_upper_estimate": 0.005306726659381202
     },
     {
       "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": 9544.56039488465,
-      "required_pages_per_second_upper_estimate": 0.003996108124252511
+      "required_token_positions_per_second": 9588.299862703143,
+      "required_pages_per_second_upper_estimate": 0.004014420926044162
     },
     {
       "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": 19089.1207897693,
-      "required_pages_per_second_upper_estimate": 0.003975760446540595
+      "required_token_positions_per_second": 19176.599725406286,
+      "required_pages_per_second_upper_estimate": 0.00399398000185911
     },
     {
       "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": 28633.681184653946,
-      "required_pages_per_second_upper_estimate": 0.0017165454659203688
+      "required_token_positions_per_second": 28764.89958810943,
+      "required_pages_per_second_upper_estimate": 0.0017244118088486243
     },
     {
       "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": 28633.681184653946,
-      "required_pages_per_second_upper_estimate": 0.0022261427658537233
+      "required_token_positions_per_second": 28764.89958810943,
+      "required_pages_per_second_upper_estimate": 0.0022363444195536274
     },
     {
       "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": 28633.681184653946,
-      "required_pages_per_second_upper_estimate": 0.36728211778522524
+      "required_token_positions_per_second": 28764.89958810943,
+      "required_pages_per_second_upper_estimate": 0.3689652465734076
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12915214557023091,
-    "producer_p95_seconds": 0.29014471999835223,
-    "gpu_update_seconds": 0.9655761626213043,
-    "producer_batches_per_second": 7.7428059408909755,
-    "gpu_batches_per_second": 1.0356510845143934,
-    "producer_headroom": 7.476268848326946,
+    "producer_batch_seconds": 0.3903516224217469,
+    "producer_p95_seconds": 0.27601766799853067,
+    "gpu_update_seconds": 0.9611714414406952,
+    "producer_batches_per_second": 2.5617928620252326,
+    "gpu_batches_per_second": 1.0403971205190041,
+    "producer_headroom": 2.462322137865277,
     "sustainable": true,
-    "buffer_seconds": 60.83129824514217,
+    "buffer_seconds": 60.553800810763796,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416770.5947933443,
-    "upload_bytes_per_second": 4858541.665841039,
+    "generation_bytes_per_second": 372069.62757039926,
+    "upload_bytes_per_second": 4900945.078934551,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.429267719748,
+    "estimated_full_shards_per_hour": 79.83748073896392,
     "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": 3.1717748339979153,
-      "bytes": 13032899,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.520787964000192,
       "bytes": 10348218,
@@ -562,6 +556,12 @@
       "bytes": 2413023,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0274238659985713,
+      "bytes": 6279174,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -276,3 +276,4 @@
 {"step": 787600, "loss": 0.5598794221878052, "lr": 3e-05, "grad_norm": 0.2708024423702626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115282045369, "tokens_per_second": 151219.01642727273, "data_wait_s": 1.0507003025850281e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 787700, "loss": 0.5161257386207581, "lr": 3e-05, "grad_norm": 0.27035623433778305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115296733947, "tokens_per_second": 152923.07279877606, "data_wait_s": 1.2881999282399192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 787800, "loss": 0.5643776059150696, "lr": 3e-05, "grad_norm": 0.2728983841707396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115311513899, "tokens_per_second": 153320.49250488, "data_wait_s": 9.920000593410805e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 787900, "loss": 0.5382817387580872, "lr": 3e-05, "grad_norm": 0.255097485272613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115326185248, "tokens_per_second": 152130.95550200183, "data_wait_s": 1.150599928223528e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5aa327a7e6eeff0e8cc0e339f8e1befc4e90084b12f3a402cf9699c5aa222272
-size 592884
+oid sha256:b8c0c6adae8d1111d08589a6f591ea0bb535e719c4f19ea2d4117f4826cb0b52
+size 595002