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": 76452.01994847241,
-      "required_pages_per_second_upper_estimate": 0.005289128133161606
+      "required_token_positions_per_second": 77400.25000307089,
+      "required_pages_per_second_upper_estimate": 0.005354728888535582
     },
     {
       "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": 9556.502493559052,
-      "required_pages_per_second_upper_estimate": 0.004001108031588105
+      "required_token_positions_per_second": 9675.031250383861,
+      "required_pages_per_second_upper_estimate": 0.00405073354690875
     },
     {
       "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": 19113.004987118104,
-      "required_pages_per_second_upper_estimate": 0.003980734894979711
+      "required_token_positions_per_second": 19350.062500767723,
+      "required_pages_per_second_upper_estimate": 0.004030107723445888
     },
     {
       "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": 28669.507480677155,
-      "required_pages_per_second_upper_estimate": 0.0017186931976633736
+      "required_token_positions_per_second": 29025.093751151584,
+      "required_pages_per_second_upper_estimate": 0.0017400100516295185
     },
     {
       "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": 28669.507480677155,
-      "required_pages_per_second_upper_estimate": 0.0022289281027863058
+      "required_token_positions_per_second": 29025.093751151584,
+      "required_pages_per_second_upper_estimate": 0.0022565733712569835
     },
     {
       "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": 28669.507480677155,
-      "required_pages_per_second_upper_estimate": 0.36774165904333134
+      "required_token_positions_per_second": 29025.093751151584,
+      "required_pages_per_second_upper_estimate": 0.37230273792218654
     }
   ],
   "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": 0.3321738296640433,
-    "producer_p95_seconds": 0.29735259199878783,
-    "gpu_update_seconds": 0.964369549028157,
-    "producer_batches_per_second": 3.010471959851227,
-    "gpu_batches_per_second": 1.0369468851518069,
-    "producer_headroom": 2.9032074862836397,
+    "producer_batch_seconds": 0.35913388596105733,
+    "producer_p95_seconds": 0.2784988120001799,
+    "gpu_update_seconds": 0.9525550627688516,
+    "producer_batches_per_second": 2.7844768736427032,
+    "gpu_batches_per_second": 1.0498080783836656,
+    "producer_headroom": 2.652367543151141,
     "sustainable": true,
-    "buffer_seconds": 60.75528158877389,
+    "buffer_seconds": 60.01096895443765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337523.4591588859,
-    "upload_bytes_per_second": 4773346.811410508,
+    "generation_bytes_per_second": 324723.77661792416,
+    "upload_bytes_per_second": 4820608.991968431,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.42467719149525,
+    "estimated_full_shards_per_hour": 69.6781632795648,
     "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": 1.2658942909984034,
-      "bytes": 6109615,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.894670263001899,
       "bytes": 6431784,
@@ -562,6 +556,12 @@
       "bytes": 6059042,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.5521136860006663,
+      "bytes": 11684886,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -310,3 +310,4 @@
 {"step": 791000, "loss": 0.507858395576477, "lr": 3e-05, "grad_norm": 0.2734002704141276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115779772392, "tokens_per_second": 152049.74784515027, "data_wait_s": 1.0251002095174044e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 791100, "loss": 0.4920814335346222, "lr": 3e-05, "grad_norm": 0.2595631820435781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115794328226, "tokens_per_second": 152004.21643786854, "data_wait_s": 1.529400105937384e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 791200, "loss": 0.5363965630531311, "lr": 3e-05, "grad_norm": 0.278750405515782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115808961492, "tokens_per_second": 151780.25353076376, "data_wait_s": 1.0931002179859206e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 791300, "loss": 0.5309306979179382, "lr": 3e-05, "grad_norm": 0.26792184818739084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115823612893, "tokens_per_second": 151635.26995532692, "data_wait_s": 1.5662000805605203e-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 sha256:6e2a900181d79e69bc4cd78b6404ceb4d8ddf7ab4125e3aefdc432696a82123e
-size 665679
+oid sha256:6bc2c6854468e8af443343e01e57d752203c4915bc9c1ac6c72726ba730d67da
+size 667797