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": 77509.4135665942,
-      "required_pages_per_second_upper_estimate": 0.005362281077154489
+      "required_token_positions_per_second": 77074.91997824048,
+      "required_pages_per_second_upper_estimate": 0.005332221802548156
     },
     {
       "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": 9688.676695824275,
-      "required_pages_per_second_upper_estimate": 0.004056446609965348
+      "required_token_positions_per_second": 9634.36499728006,
+      "required_pages_per_second_upper_estimate": 0.00403370743594212
     },
     {
       "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": 19377.35339164855,
-      "required_pages_per_second_upper_estimate": 0.004035791696307173
+      "required_token_positions_per_second": 19268.72999456012,
+      "required_pages_per_second_upper_estimate": 0.004013168307285269
     },
     {
       "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": 29066.030087472827,
-      "required_pages_per_second_upper_estimate": 0.0017424641224858105
+      "required_token_positions_per_second": 28903.09499184018,
+      "required_pages_per_second_upper_estimate": 0.001732696412290119
     },
     {
       "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": 29066.030087472827,
-      "required_pages_per_second_upper_estimate": 0.002259755991346046
+      "required_token_positions_per_second": 28903.09499184018,
+      "required_pages_per_second_upper_estimate": 0.0022470885043363526
     },
     {
       "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": 29066.030087472827,
-      "required_pages_per_second_upper_estimate": 0.372827825290502
+      "required_token_positions_per_second": 28903.09499184018,
+      "required_pages_per_second_upper_estimate": 0.3707378688298018
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07261075934354722,
-    "producer_p95_seconds": 0.25364185299986275,
-    "gpu_update_seconds": 0.951213492754073,
-    "producer_batches_per_second": 13.772063658894488,
-    "gpu_batches_per_second": 1.0512887039739882,
-    "producer_headroom": 13.100172775408463,
+    "producer_batch_seconds": 0.2785931907812653,
+    "producer_p95_seconds": 0.30296316200110596,
+    "gpu_update_seconds": 0.9565757579873535,
+    "producer_batches_per_second": 3.5894631781763113,
+    "gpu_batches_per_second": 1.0453955075173675,
+    "producer_headroom": 3.4335934604316996,
     "sustainable": true,
-    "buffer_seconds": 60.87766353626067,
+    "buffer_seconds": 60.26427275320327,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 32449.37928566523,
-    "upload_bytes_per_second": 5087313.323269238,
+    "generation_bytes_per_second": 32698.17773124006,
+    "upload_bytes_per_second": 4960235.699368814,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 6.962881411814381,
+    "estimated_full_shards_per_hour": 7.016267766503347,
     "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": 1.3705294199971831,
-      "bytes": 12626897,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.047090729996853,
       "bytes": 5981571,
@@ -562,6 +556,12 @@
       "bytes": 6889170,
       "pages": 43,
       "shards": 2
+    },
+    {
+      "seconds": 0.9624333549982111,
+      "bytes": 484343,
+      "pages": 14,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -259,3 +259,4 @@
 {"step": 710900, "loss": 0.5743618011474609, "lr": 8e-05, "grad_norm": 0.21989600077509383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104055896511, "tokens_per_second": 43432.19885658717, "data_wait_s": 1.4179997378960252e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 711000, "loss": 0.5434914827346802, "lr": 8e-05, "grad_norm": 0.23269289903401955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104070684528, "tokens_per_second": 154419.78689260912, "data_wait_s": 1.8021000869339332e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 711100, "loss": 0.5039126873016357, "lr": 8e-05, "grad_norm": 0.21354127733016204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104085131119, "tokens_per_second": 85428.956407359, "data_wait_s": 1.4512999769067392e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 711200, "loss": 0.5821572542190552, "lr": 8e-05, "grad_norm": 0.21605494536219572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104099538817, "tokens_per_second": 149477.3624187527, "data_wait_s": 1.6649999452056363e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f55b5fc8410e55092f20a48af2c4934685235bdf94a4895af49fee74dd906551
-size 556878
+oid sha256:cb2ce00df2883850cf730d1b3103225666f5d66348bb4f9c5d7aa837682b552b
+size 558996