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": 76736.30222281809,
-      "required_pages_per_second_upper_estimate": 0.005308795440526594
+      "required_token_positions_per_second": 77055.39053682695,
+      "required_pages_per_second_upper_estimate": 0.005330870710476621
     },
     {
       "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": 9592.037777852262,
-      "required_pages_per_second_upper_estimate": 0.004015985913060553
+      "required_token_positions_per_second": 9631.923817103368,
+      "required_pages_per_second_upper_estimate": 0.004032685364779779
     },
     {
       "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": 19184.075555704523,
-      "required_pages_per_second_upper_estimate": 0.00399553702015933
+      "required_token_positions_per_second": 19263.847634206737,
+      "required_pages_per_second_upper_estimate": 0.004012151440380161
     },
     {
       "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": 28776.113333556787,
-      "required_pages_per_second_upper_estimate": 0.0017250840557657954
+      "required_token_positions_per_second": 28895.771451310105,
+      "required_pages_per_second_upper_estimate": 0.0017322573772177402
     },
     {
       "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": 28776.113333556787,
-      "required_pages_per_second_upper_estimate": 0.002237216238938106
+      "required_token_positions_per_second": 28895.771451310105,
+      "required_pages_per_second_upper_estimate": 0.0022465191312729896
     },
     {
       "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": 28776.113333556787,
-      "required_pages_per_second_upper_estimate": 0.36910908445962454
+      "required_token_positions_per_second": 28895.771451310105,
+      "required_pages_per_second_upper_estimate": 0.3706439303152872
     }
   ],
   "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": 0.05418472578124067,
-    "producer_p95_seconds": 0.17214284399960889,
-    "gpu_update_seconds": 0.960796883148175,
-    "producer_batches_per_second": 18.455385453776913,
-    "gpu_batches_per_second": 1.0408027102704278,
-    "producer_headroom": 17.731876821287027,
+    "producer_batch_seconds": 0.07241884508592733,
+    "producer_p95_seconds": 0.1788760009994803,
+    "gpu_update_seconds": 0.9568181990429768,
+    "producer_batches_per_second": 13.808560448781906,
+    "gpu_batches_per_second": 1.0451306225155566,
+    "producer_headroom": 13.212281939979583,
     "sustainable": true,
-    "buffer_seconds": 60.530203638335024,
+    "buffer_seconds": 60.27954653970754,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460014.61391499446,
+    "generation_bytes_per_second": 462273.62131834397,
     "upload_bytes_per_second": 5452492.670439877,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.70842755401016,
+    "estimated_full_shards_per_hour": 99.193157955768,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -58,3 +58,4 @@
 {"step": 610200, "loss": 0.549430251121521, "lr": 0.0001, "grad_norm": 0.16628906229233595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89315504557, "tokens_per_second": 150681.60285144622, "data_wait_s": 1.5778000488353427e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610300, "loss": 0.6092767119407654, "lr": 0.0001, "grad_norm": 0.14803397781210065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89330207406, "tokens_per_second": 152671.29239890774, "data_wait_s": 1.3412999578577e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610400, "loss": 0.5435611009597778, "lr": 0.0001, "grad_norm": 0.15360692895727052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89344914056, "tokens_per_second": 152009.4576179536, "data_wait_s": 1.6773999959696084e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 610500, "loss": 0.6074555516242981, "lr": 0.0001, "grad_norm": 0.169522746046445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89359602774, "tokens_per_second": 152499.79723854814, "data_wait_s": 1.5468999663426075e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8cdebeee24139cfe4f8bbe0d85634e81b8c701bb4ab25072d1fac1f073c5cb39
-size 128811
+oid sha256:5b33d04bf708fbcd23bcd5a1898cc0adaddb6faeb4fb927f0b4e182443461e98
+size 130929