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": 76448.44048059934,
-      "required_pages_per_second_upper_estimate": 0.005288880497268636
+      "required_token_positions_per_second": 76261.21962577773,
+      "required_pages_per_second_upper_estimate": 0.005275928124119844
     },
     {
       "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": 9556.055060074917,
-      "required_pages_per_second_upper_estimate": 0.004000920700531771
+      "required_token_positions_per_second": 9532.652453222216,
+      "required_pages_per_second_upper_estimate": 0.003991122517744548
     },
     {
       "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": 19112.110120149835,
-      "required_pages_per_second_upper_estimate": 0.003980548517789449
+      "required_token_positions_per_second": 19065.30490644443,
+      "required_pages_per_second_upper_estimate": 0.003970800226111105
     },
     {
       "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": 28668.16518022475,
-      "required_pages_per_second_upper_estimate": 0.0017186127288008261
+      "required_token_positions_per_second": 28597.957359666645,
+      "required_pages_per_second_upper_estimate": 0.0017144038771595026
     },
     {
       "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": 28668.16518022475,
-      "required_pages_per_second_upper_estimate": 0.002228823744830287
+      "required_token_positions_per_second": 28597.957359666645,
+      "required_pages_per_second_upper_estimate": 0.0022233653955934605
     },
     {
       "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": 28668.16518022475,
-      "required_pages_per_second_upper_estimate": 0.3677244414543778
+      "required_token_positions_per_second": 28597.957359666645,
+      "required_pages_per_second_upper_estimate": 0.36682389091554296
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.29027496763308136,
-    "producer_p95_seconds": 0.3467184410037589,
-    "gpu_update_seconds": 0.9644147027264772,
-    "producer_batches_per_second": 3.4450094272821974,
-    "gpu_batches_per_second": 1.0368983355116013,
-    "producer_headroom": 3.322417742702272,
+    "producer_batch_seconds": 0.11464789210941717,
+    "producer_p95_seconds": 0.33563471299567027,
+    "gpu_update_seconds": 0.9667823352654401,
+    "producer_batches_per_second": 8.722358358282106,
+    "gpu_batches_per_second": 1.03435899014998,
+    "producer_headroom": 8.432621982642004,
     "sustainable": true,
-    "buffer_seconds": 60.75812627176806,
+    "buffer_seconds": 60.907287121722725,
     "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": 420999.86127386946,
-    "upload_bytes_per_second": 5528497.713898112,
+    "generation_bytes_per_second": 438304.62170390435,
+    "upload_bytes_per_second": 5641860.546302463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.3367698541838,
+    "estimated_full_shards_per_hour": 94.04996860826347,
     "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": 1.0836265130019456,
-      "bytes": 2657150,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 0.9289632090003579,
       "bytes": 5852982,
@@ -562,6 +556,12 @@
       "bytes": 1026854,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.3134237790000043,
+      "bytes": 12455748,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -374,3 +374,4 @@
 {"step": 722400, "loss": 0.5172302722930908, "lr": 8e-05, "grad_norm": 0.21287814167332977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105739363686, "tokens_per_second": 154658.96207514982, "data_wait_s": 1.5737001376692206e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 722500, "loss": 0.519875705242157, "lr": 8e-05, "grad_norm": 0.22840929185045208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105754059450, "tokens_per_second": 151841.19993489745, "data_wait_s": 1.5067998901940882e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 722600, "loss": 0.5492111444473267, "lr": 8e-05, "grad_norm": 0.22936311170345755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105768734296, "tokens_per_second": 152930.9104988477, "data_wait_s": 1.4805998944211751e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 722700, "loss": 0.5891305804252625, "lr": 8e-05, "grad_norm": 0.23494695261280726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105783284226, "tokens_per_second": 151962.07370628853, "data_wait_s": 9.780000254977494e-06, "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 sha256:4d9a3cc2dc30e9c69998ad133e2322bbecb13b191d87c8b47b34ca448437d191
-size 802014
+oid sha256:2af9e8e6c680f9cadffc7076a1635de288e1bf9db0fd3c9188a844e0fde7d19a
+size 804132