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": 75695.36437790014,
-      "required_pages_per_second_upper_estimate": 0.006561082336129329
+      "required_token_positions_per_second": 75672.96753388397,
+      "required_pages_per_second_upper_estimate": 0.00655914103445429
     },
     {
       "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": 9461.920547237518,
-      "required_pages_per_second_upper_estimate": 0.0037745015746120623
+      "required_token_positions_per_second": 9459.120941735497,
+      "required_pages_per_second_upper_estimate": 0.0037733847701194737
     },
     {
       "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": 18923.841094475036,
-      "required_pages_per_second_upper_estimate": 0.004377697588978546
+      "required_token_positions_per_second": 18918.241883470993,
+      "required_pages_per_second_upper_estimate": 0.004376402310055509
     },
     {
       "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": 28385.761641712554,
-      "required_pages_per_second_upper_estimate": 0.0017285439553987676
+      "required_token_positions_per_second": 28377.362825206492,
+      "required_pages_per_second_upper_estimate": 0.0017280325115387333
     },
     {
       "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": 28385.761641712554,
-      "required_pages_per_second_upper_estimate": 0.0021852145079963967
+      "required_token_positions_per_second": 28377.362825206492,
+      "required_pages_per_second_upper_estimate": 0.002184567943852348
     },
     {
       "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": 28385.761641712554,
-      "required_pages_per_second_upper_estimate": 0.35245615855709245
+      "required_token_positions_per_second": 28377.362825206492,
+      "required_pages_per_second_upper_estimate": 0.3523518733651178
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10079627726554463,
-    "producer_p95_seconds": 0.30353833100161864,
-    "gpu_update_seconds": 0.974009447024017,
-    "producer_batches_per_second": 9.92100132195886,
-    "gpu_batches_per_second": 1.0266840871568488,
-    "producer_headroom": 9.663149011525691,
+    "producer_batch_seconds": 0.1275587699061589,
+    "producer_p95_seconds": 0.34824691000176244,
+    "gpu_update_seconds": 0.9742977235164845,
+    "producer_batches_per_second": 7.839523701394029,
+    "gpu_batches_per_second": 1.0263803105181746,
+    "producer_headroom": 7.638030095721727,
     "sustainable": true,
-    "buffer_seconds": 60.38858571548906,
+    "buffer_seconds": 60.406458858022035,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336347.3401759037,
-    "upload_bytes_per_second": 3331568.863543371,
+    "generation_bytes_per_second": 346306.9395944365,
+    "upload_bytes_per_second": 3249932.2128365957,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.17230943639596,
+    "estimated_full_shards_per_hour": 74.30940762400456,
     "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": 1.0075569920009002,
-      "bytes": 5909692,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 0.7540812590013957,
       "bytes": 2508470,
@@ -562,6 +556,12 @@
       "bytes": 822769,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 4.258936992002418,
+      "bytes": 6104340,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -286,3 +286,4 @@
 {"step": 68600, "loss": 0.604379765689373, "lr": 0.0003, "grad_norm": 0.13055077179418748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10039940722, "tokens_per_second": 147720.23354195536, "data_wait_s": 1.296499976888299e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 68700, "loss": 0.5999638177454472, "lr": 0.0003, "grad_norm": 0.13280039697618273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10054465618, "tokens_per_second": 148837.63019668384, "data_wait_s": 1.3896999007556587e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 68800, "loss": 0.6019186750054359, "lr": 0.0003, "grad_norm": 0.13937680307974623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10069024094, "tokens_per_second": 147734.42560785532, "data_wait_s": 9.390001650899649e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 68900, "loss": 0.6303941160440445, "lr": 0.0003, "grad_norm": 0.13800357157853751, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10083778947, "tokens_per_second": 149816.38426153688, "data_wait_s": 1.1046999134123325e-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 sha256:4882c05064b014f4e852e1df1ac093adf74d575cfd7e00243525456c64534987
-size 614064
+oid sha256:02767cccfb24625f50ee825354accf13bca81d56c76bce6685fbc06a254d6dc2
+size 616182