Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76243.43453323765,
-      "required_pages_per_second_upper_estimate": 0.006608587668122876
+      "required_token_positions_per_second": 76192.36791717181,
+      "required_pages_per_second_upper_estimate": 0.006604161343269444
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9530.429316654707,
-      "required_pages_per_second_upper_estimate": 0.0038018307470299613
+      "required_token_positions_per_second": 9524.045989646476,
+      "required_pages_per_second_upper_estimate": 0.0037992843424471344
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19060.858633309414,
-      "required_pages_per_second_upper_estimate": 0.004409394185161557
+      "required_token_positions_per_second": 19048.091979292953,
+      "required_pages_per_second_upper_estimate": 0.004406440844440285
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28591.28794996412,
-      "required_pages_per_second_upper_estimate": 0.0017410594292580851
+      "required_token_positions_per_second": 28572.13796893943,
+      "required_pages_per_second_upper_estimate": 0.001739893295885867
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28591.28794996412,
-      "required_pages_per_second_upper_estimate": 0.002201036492138837
+      "required_token_positions_per_second": 28572.13796893943,
+      "required_pages_per_second_upper_estimate": 0.002199562273589016
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28591.28794996412,
-      "required_pages_per_second_upper_estimate": 0.35500810745327144
+      "required_token_positions_per_second": 28572.13796893943,
+      "required_pages_per_second_upper_estimate": 0.3547703287798084
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.13947011276550825,
-    "producer_p95_seconds": 0.22885370699987106,
-    "gpu_update_seconds": 0.967007853874407,
-    "producer_batches_per_second": 7.169994919852862,
-    "gpu_batches_per_second": 1.034117764393957,
-    "producer_headroom": 6.933441399737317,
+    "producer_batch_seconds": 0.0764171929608608,
+    "producer_p95_seconds": 0.19693693299996085,
+    "gpu_update_seconds": 0.9676559741541199,
+    "producer_batches_per_second": 13.08606036487336,
+    "gpu_batches_per_second": 1.0334251290849041,
+    "producer_headroom": 12.662804490211148,
     "sustainable": true,
-    "buffer_seconds": 60.92149479408764,
+    "buffer_seconds": 60.96232637170955,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 496799.57260196295,
-    "upload_bytes_per_second": 6089034.86801216,
+    "generation_bytes_per_second": 499501.9404014323,
+    "upload_bytes_per_second": 6015733.198680412,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.60162337822119,
+    "estimated_full_shards_per_hour": 107.18148860008455,
     "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": 1.141071416000159,
-      "bytes": 10868715,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.1746170310007074,
       "bytes": 7476573,
@@ -562,6 +556,12 @@
       "bytes": 7538580,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1064888179989794,
+      "bytes": 5541578,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -120,3 +120,4 @@
 {"step": 52000, "loss": 0.6502221059054136, "lr": 0.0003, "grad_norm": 0.15132006725584615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7610525400, "tokens_per_second": 147422.74991536338, "data_wait_s": 9.97200004348997e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 52100, "loss": 0.5977641493082047, "lr": 0.0003, "grad_norm": 0.1296948953688188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7625174466, "tokens_per_second": 150817.8264087898, "data_wait_s": 1.3049999324721284e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 52200, "loss": 0.6357814688235521, "lr": 0.0003, "grad_norm": 0.14869620170789155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7639880011, "tokens_per_second": 151184.7904440597, "data_wait_s": 6.382999345078133e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 52300, "loss": 0.6685989703983068, "lr": 0.0003, "grad_norm": 0.1285756936707822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7654581453, "tokens_per_second": 149826.61015054796, "data_wait_s": 1.7459000446251594e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f7a7d60a0891ba3e1216d569a17821a152026f3ff6ca33340da4a7f5e323a1d2
-size 260127
+oid sha256:079bc118f6bffdb3a47dfd7220f30bb6c50f84b6947a39821acf3738bb4ced8d
+size 262245