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": 75150.71452245767,
-      "required_pages_per_second_upper_estimate": 0.006513873467062014
+      "required_token_positions_per_second": 74535.89493664996,
+      "required_pages_per_second_upper_estimate": 0.006460582463610196
     },
     {
       "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": 9393.839315307208,
-      "required_pages_per_second_upper_estimate": 0.0037473429532899347
+      "required_token_positions_per_second": 9316.986867081245,
+      "required_pages_per_second_upper_estimate": 0.0037166853626461006
     },
     {
       "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": 18787.678630614417,
-      "required_pages_per_second_upper_estimate": 0.004346198799341909
+      "required_token_positions_per_second": 18633.97373416249,
+      "required_pages_per_second_upper_estimate": 0.0043106418766615424
     },
     {
       "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": 28181.517945921623,
-      "required_pages_per_second_upper_estimate": 0.0017161065859089555
+      "required_token_positions_per_second": 27950.960601243733,
+      "required_pages_per_second_upper_estimate": 0.0017020668532589736
     },
     {
       "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": 28181.517945921623,
-      "required_pages_per_second_upper_estimate": 0.002169491262911675
+      "required_token_positions_per_second": 27950.960601243733,
+      "required_pages_per_second_upper_estimate": 0.0021517423202947927
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28181.517945921623,
-      "required_pages_per_second_upper_estimate": 0.3499201354150468
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27950.960601243733,
+      "required_pages_per_second_upper_estimate": 0.3470573848199428
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.14266494116412787,
-    "producer_p95_seconds": 0.3790997399992193,
-    "gpu_update_seconds": 0.9810685163607792,
-    "producer_batches_per_second": 7.009430570959666,
-    "gpu_batches_per_second": 1.0192968007060772,
-    "producer_headroom": 6.876731650785289,
+    "producer_batch_seconds": 0.5422224882499904,
+    "producer_p95_seconds": 0.3886906879997696,
+    "gpu_update_seconds": 0.9891609950167418,
+    "producer_batches_per_second": 1.8442613902412552,
+    "gpu_batches_per_second": 1.0109577763760031,
+    "producer_headroom": 1.8242714318419997,
     "sustainable": true,
-    "buffer_seconds": 60.82624801436831,
+    "buffer_seconds": 60.33882069602125,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 317949.0612690121,
-    "upload_bytes_per_second": 5238277.144169608,
+    "generation_bytes_per_second": 269086.28562573297,
+    "upload_bytes_per_second": 5270144.002100331,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.22446707299015,
+    "estimated_full_shards_per_hour": 57.73965288714401,
     "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": 1.2693092910012638,
-      "bytes": 6119669,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.0424042099984945,
       "bytes": 2508707,
@@ -562,6 +556,12 @@
       "bytes": 6199366,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0861811289942125,
+      "bytes": 7695547,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -343,3 +343,4 @@
 {"step": 389300, "loss": 0.6070778537541628, "lr": 0.0003, "grad_norm": 0.14442209134593403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56983098103, "tokens_per_second": 150553.5773020793, "data_wait_s": 1.2715005141217262e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 389400, "loss": 0.5318893380463123, "lr": 0.0003, "grad_norm": 0.1610027918707639, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56997768709, "tokens_per_second": 148459.80504129443, "data_wait_s": 1.5739002265036106e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 389500, "loss": 0.5847944058477879, "lr": 0.0003, "grad_norm": 0.1547880307632339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57012298598, "tokens_per_second": 138880.10867600585, "data_wait_s": 1.112099562305957e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 389600, "loss": 0.5617741234600544, "lr": 0.0003, "grad_norm": 0.14156084775089878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57026864321, "tokens_per_second": 128063.9472956184, "data_wait_s": 1.4544995792675763e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:19f2967ca2360d05a5857724ede6bededf95a755b9619301fafe0e897849dbb0
-size 735573
+oid sha256:76c914bc394a49dc6db7661f5337517487f97d990715ab50ebf6e6b72ecbf637
+size 737691