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": 75568.6064213642,
-      "required_pages_per_second_upper_estimate": 0.006550095277721897
+      "required_token_positions_per_second": 75408.98840548929,
+      "required_pages_per_second_upper_estimate": 0.006536259992654023
     },
     {
       "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": 9446.075802670524,
-      "required_pages_per_second_upper_estimate": 0.0037681808691042463
+      "required_token_positions_per_second": 9426.123550686161,
+      "required_pages_per_second_upper_estimate": 0.0037602216174749325
     },
     {
       "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": 18892.15160534105,
-      "required_pages_per_second_upper_estimate": 0.004370366783383359
+      "required_token_positions_per_second": 18852.247101372323,
+      "required_pages_per_second_upper_estimate": 0.0043611355786854775
     },
     {
       "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": 28338.22740801157,
-      "required_pages_per_second_upper_estimate": 0.001725649369959228
+      "required_token_positions_per_second": 28278.370652058482,
+      "required_pages_per_second_upper_estimate": 0.0017220044075658128
     },
     {
       "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": 28338.22740801157,
-      "required_pages_per_second_upper_estimate": 0.0021815551911029147
+      "required_token_positions_per_second": 28278.370652058482,
+      "required_pages_per_second_upper_estimate": 0.0021769472523355403
     },
     {
       "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": 28338.22740801157,
-      "required_pages_per_second_upper_estimate": 0.35186594246137265
+      "required_token_positions_per_second": 28278.370652058482,
+      "required_pages_per_second_upper_estimate": 0.3511227218801108
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.07046789303125678,
-    "producer_p95_seconds": 0.17226098599985562,
-    "gpu_update_seconds": 0.9756432398514653,
-    "producer_batches_per_second": 14.190859936119269,
-    "gpu_batches_per_second": 1.0249648223383814,
-    "producer_headroom": 13.84521656435376,
+    "producer_batch_seconds": 0.07957594310155969,
+    "producer_p95_seconds": 0.16779538700029661,
+    "gpu_update_seconds": 0.9777083814405483,
+    "producer_batches_per_second": 12.566611981258442,
+    "gpu_batches_per_second": 1.0227998644407725,
+    "producer_headroom": 12.286481860387592,
     "sustainable": true,
-    "buffer_seconds": 60.48988087079085,
+    "buffer_seconds": 60.61791964931399,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443125.1305652237,
-    "upload_bytes_per_second": 5258659.707611809,
+    "generation_bytes_per_second": 441772.35945116676,
+    "upload_bytes_per_second": 5295845.539422616,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.0843375941995,
+    "estimated_full_shards_per_hour": 94.79406440402272,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 5661667,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0493596160004017,
+      "bytes": 7456331,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -51,3 +51,4 @@
 {"step": 320100, "loss": 0.6229025758802891, "lr": 0.0003, "grad_norm": 0.1792688815400863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46853753312, "tokens_per_second": 131315.68225631822, "data_wait_s": 1.069899917638395e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 320200, "loss": 0.5868458636105061, "lr": 0.0003, "grad_norm": 0.14736664657288578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46868391316, "tokens_per_second": 148882.3515940924, "data_wait_s": 9.066000529855955e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 320300, "loss": 0.5900352392345667, "lr": 0.0003, "grad_norm": 0.1685965670996436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46883153945, "tokens_per_second": 149671.4621128526, "data_wait_s": 1.3650000255438499e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 320400, "loss": 0.5965177156031132, "lr": 0.0003, "grad_norm": 0.16110717642181044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46897838392, "tokens_per_second": 148941.93556122648, "data_wait_s": 1.3484000191965606e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:23816e2f29dccf156542c7c085f5723cac64ca01baa57409e86a22cda2f637d6
-size 113202
+oid sha256:15d6edce58d08f74ebaf46c6c8aaadfaed2f032576bd860cdabf5ad64d1b2326
+size 115320