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": 77272.13278922167,
-      "required_pages_per_second_upper_estimate": 0.005345865442408601
+      "required_token_positions_per_second": 77005.829375506,
+      "required_pages_per_second_upper_estimate": 0.0053274419543387
     },
     {
       "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": 9659.016598652708,
-      "required_pages_per_second_upper_estimate": 0.004044028546652873
+      "required_token_positions_per_second": 9625.72867193825,
+      "required_pages_per_second_upper_estimate": 0.004030091586868486
     },
     {
       "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": 19318.033197305416,
-      "required_pages_per_second_upper_estimate": 0.004023436864204223
+      "required_token_positions_per_second": 19251.4573438765,
+      "required_pages_per_second_upper_estimate": 0.0040095708696583025
     },
     {
       "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": 28977.049795958123,
-      "required_pages_per_second_upper_estimate": 0.0017371298898745423
+      "required_token_positions_per_second": 28877.18601581475,
+      "required_pages_per_second_upper_estimate": 0.0017311432087381104
     },
     {
       "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": 28977.049795958123,
-      "required_pages_per_second_upper_estimate": 0.0022528381650637175
+      "required_token_positions_per_second": 28877.18601581475,
+      "required_pages_per_second_upper_estimate": 0.0022450741954119172
     },
     {
       "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": 28977.049795958123,
-      "required_pages_per_second_upper_estimate": 0.3716864816505448
+      "required_token_positions_per_second": 28877.18601581475,
+      "required_pages_per_second_upper_estimate": 0.37040553630423867
     }
   ],
   "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": 0.1293164644997944,
-    "producer_p95_seconds": 0.25933782199717825,
-    "gpu_update_seconds": 0.9541343992809266,
-    "producer_batches_per_second": 7.732967367056265,
-    "gpu_batches_per_second": 1.048070377457976,
-    "producer_headroom": 7.3782901734252375,
+    "producer_batch_seconds": 0.12785020585934603,
+    "producer_p95_seconds": 0.2753374919993803,
+    "gpu_update_seconds": 0.9574340098394082,
+    "producer_batches_per_second": 7.821653420723832,
+    "gpu_batches_per_second": 1.0444584062432998,
+    "producer_headroom": 7.488716998177742,
     "sustainable": true,
-    "buffer_seconds": 60.11046715469838,
+    "buffer_seconds": 60.31834261988271,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 264399.37543163967,
-    "upload_bytes_per_second": 5017029.053905758,
+    "generation_bytes_per_second": 275834.8374770431,
+    "upload_bytes_per_second": 4925225.78443723,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.73395106517689,
+    "estimated_full_shards_per_hour": 59.187735016188334,
     "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": 1.4136529820007127,
-      "bytes": 11858220,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.0470604990005086,
       "bytes": 6246532,
@@ -562,6 +556,12 @@
       "bytes": 9956766,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1540077350000502,
+      "bytes": 2614009,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -210,3 +210,4 @@
 {"step": 781000, "loss": 0.5303976535797119, "lr": 3e-05, "grad_norm": 0.26523995730134536, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114316271822, "tokens_per_second": 132735.53892428303, "data_wait_s": 1.5020003047538921e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 781100, "loss": 0.5578319430351257, "lr": 3e-05, "grad_norm": 0.2528055753867834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114330886818, "tokens_per_second": 152319.48281221697, "data_wait_s": 1.2610002158908173e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 781200, "loss": 0.5123441815376282, "lr": 3e-05, "grad_norm": 0.24898188702796165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114345432347, "tokens_per_second": 150619.8920894047, "data_wait_s": 1.2929001968586817e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 781300, "loss": 0.5730502009391785, "lr": 3e-05, "grad_norm": 0.25821237201401737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114360065856, "tokens_per_second": 151143.15535135104, "data_wait_s": 9.031999070430174e-06, "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 sha256:55407457b8e751d9ed442b8d6c476a2b803ef26c7374b3cd62e3c7ed882f7cec
-size 452313
+oid sha256:8eda8de2e777b0b0b640b0de195c902e65eef052f427ab82c358c9754d42862b
+size 454431