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": 76594.44717119788,
-      "required_pages_per_second_upper_estimate": 0.0052989815789064405
+      "required_token_positions_per_second": 77100.81776463869,
+      "required_pages_per_second_upper_estimate": 0.005334013471502328
     },
     {
       "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": 9574.305896399736,
-      "required_pages_per_second_upper_estimate": 0.004008561944580172
+      "required_token_positions_per_second": 9637.602220579836,
+      "required_pages_per_second_upper_estimate": 0.004035062793736827
     },
     {
       "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": 19148.61179279947,
-      "required_pages_per_second_upper_estimate": 0.003988150853588527
+      "required_token_positions_per_second": 19275.204441159673,
+      "required_pages_per_second_upper_estimate": 0.004014516763769318
     },
     {
       "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": 28722.917689199207,
-      "required_pages_per_second_upper_estimate": 0.0017218950581116043
+      "required_token_positions_per_second": 28912.80666173951,
+      "required_pages_per_second_upper_estimate": 0.0017332786120717227
     },
     {
       "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": 28722.917689199207,
-      "required_pages_per_second_upper_estimate": 0.0022330805115722157
+      "required_token_positions_per_second": 28912.80666173951,
+      "required_pages_per_second_upper_estimate": 0.0022478435439539043
     },
     {
       "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": 28722.917689199207,
-      "required_pages_per_second_upper_estimate": 0.36842674785083834
+      "required_token_positions_per_second": 28912.80666173951,
+      "required_pages_per_second_upper_estimate": 0.3708624397036917
     }
   ],
   "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": 0.19462399756241666,
-    "producer_p95_seconds": 0.33406189599918434,
-    "gpu_update_seconds": 0.9625763057628574,
-    "producer_batches_per_second": 5.1381125273582775,
-    "gpu_batches_per_second": 1.038878677994763,
-    "producer_headroom": 4.94582537517839,
+    "producer_batch_seconds": 0.10756448981237554,
+    "producer_p95_seconds": 0.31914502599829575,
+    "gpu_update_seconds": 0.9562544488836071,
+    "producer_batches_per_second": 9.296748413387144,
+    "gpu_batches_per_second": 1.0457467687261108,
+    "producer_headroom": 8.890057030453073,
     "sustainable": true,
-    "buffer_seconds": 60.64230726306001,
+    "buffer_seconds": 60.24403027966725,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 437342.30100081593,
+    "generation_bytes_per_second": 436481.773757183,
     "upload_bytes_per_second": 5092625.280321767,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.84347698706016,
+    "estimated_full_shards_per_hour": 93.65882787262552,
     "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

@@ -343,3 +343,4 @@
 {"step": 638700, "loss": 0.5350292921066284, "lr": 0.0001, "grad_norm": 0.16551516570421757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93489464400, "tokens_per_second": 152998.01491213532, "data_wait_s": 1.2072996469214559e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 638800, "loss": 0.5327662229537964, "lr": 0.0001, "grad_norm": 0.1819359424993719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93504217745, "tokens_per_second": 136444.39995466507, "data_wait_s": 1.4217999705579132e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 638900, "loss": 0.5622432827949524, "lr": 0.0001, "grad_norm": 0.15591757221113622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93518982482, "tokens_per_second": 153724.44244898058, "data_wait_s": 1.0813993867486715e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 639000, "loss": 0.5632246136665344, "lr": 0.0001, "grad_norm": 0.17179790947356524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93533736230, "tokens_per_second": 152623.36201162174, "data_wait_s": 1.8385995645076036e-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:caf04db9e34fbaf5117a78e38146fe450273e82b96e91da40ac932c5073b54c6
-size 736356
+oid sha256:beac0ac9b9e8e3c08f8bc5d4dd24fb5d4ad53d6495bd0aa851ebd2f9a5b16577
+size 738474