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": 73620.21806656512,
-      "required_pages_per_second_upper_estimate": 0.006381213913273053
+      "required_token_positions_per_second": 73898.6054645079,
+      "required_pages_per_second_upper_estimate": 0.006405343827360323
     },
     {
       "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": 9202.52725832064,
-      "required_pages_per_second_upper_estimate": 0.0036710257133878415
+      "required_token_positions_per_second": 9237.325683063487,
+      "required_pages_per_second_upper_estimate": 0.0036849073253005776
     },
     {
       "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": 18405.05451664128,
-      "required_pages_per_second_upper_estimate": 0.004257685444528638
+      "required_token_positions_per_second": 18474.651366126975,
+      "required_pages_per_second_upper_estimate": 0.004273785450794436
     },
     {
       "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": 27607.581774961924,
-      "required_pages_per_second_upper_estimate": 0.0016811568843078802
+      "required_token_positions_per_second": 27711.977049190464,
+      "required_pages_per_second_upper_estimate": 0.0016875140087887245
     },
     {
       "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": 27607.581774961924,
-      "required_pages_per_second_upper_estimate": 0.002125308067714191
+      "required_token_positions_per_second": 27711.977049190464,
+      "required_pages_per_second_upper_estimate": 0.0021333447049089187
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27607.581774961924,
-      "required_pages_per_second_upper_estimate": 0.34279376901252745
+      "required_token_positions_per_second": 27711.977049190464,
+      "required_pages_per_second_upper_estimate": 0.3440900089299386
     }
   ],
   "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": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13758849685126506,
-    "producer_p95_seconds": 0.3396952609982691,
-    "gpu_update_seconds": 1.001464026272476,
-    "producer_batches_per_second": 7.268049458240778,
-    "gpu_batches_per_second": 0.9985381139670835,
-    "producer_headroom": 7.278690073597298,
+    "producer_batch_seconds": 0.36286089454682724,
+    "producer_p95_seconds": 0.35293912699853536,
+    "gpu_update_seconds": 0.9976913574561317,
+    "producer_batches_per_second": 2.755877017965489,
+    "gpu_batches_per_second": 1.0023139847074096,
+    "producer_headroom": 2.749514683036145,
     "sustainable": true,
-    "buffer_seconds": 60.087841576348566,
+    "buffer_seconds": 60.859172804824034,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 515582.70211433456,
-    "upload_bytes_per_second": 5700488.494928551,
+    "generation_bytes_per_second": 461680.7093863655,
+    "upload_bytes_per_second": 5715564.254599713,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.63204572269943,
+    "estimated_full_shards_per_hour": 99.0659328574488,
     "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": 0.9942282960000739,
-      "bytes": 5730133,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9958830260002287,
       "bytes": 6097937,
@@ -562,6 +556,12 @@
       "bytes": 1005909,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.0930767579993699,
+      "bytes": 7383476,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -257,3 +257,4 @@
 {"step": 25700, "loss": 0.6349320039153099, "lr": 0.0003, "grad_norm": 0.14398465931087587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3760551023, "tokens_per_second": 148601.50113188717, "data_wait_s": 1.2879001587862149e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 25800, "loss": 0.6204517334699631, "lr": 0.0003, "grad_norm": 0.13723740660540482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3775161785, "tokens_per_second": 148692.41507626904, "data_wait_s": 7.5120005931239575e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 25900, "loss": 0.6570776477456093, "lr": 0.0003, "grad_norm": 0.14541464004758917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3789806964, "tokens_per_second": 148410.55718301475, "data_wait_s": 1.1101001291535795e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 26000, "loss": 0.6907591354101896, "lr": 0.0003, "grad_norm": 0.14134570257389986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3804476685, "tokens_per_second": 149261.4004277892, "data_wait_s": 1.2415999663062394e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1a6f93de3b09cf1291f37efcf62cd70bb7b967260ea20b9aa533862ecb343799
-size 546868
+oid sha256:6b72d9edc49361d1a25a18309483963792de7aef8e7f2aafe52897aa3de4cd25
+size 548986