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": 75228.80248315382,
-      "required_pages_per_second_upper_estimate": 0.0052045031106553105
+      "required_token_positions_per_second": 76046.6656809564,
+      "required_pages_per_second_upper_estimate": 0.005261084784383364
     },
     {
       "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": 9403.600310394228,
-      "required_pages_per_second_upper_estimate": 0.003937091080457662
+      "required_token_positions_per_second": 9505.83321011955,
+      "required_pages_per_second_upper_estimate": 0.003979893860706944
     },
     {
       "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": 18807.200620788455,
-      "required_pages_per_second_upper_estimate": 0.003917043910076706
+      "required_token_positions_per_second": 19011.6664202391,
+      "required_pages_per_second_upper_estimate": 0.003959628743976539
     },
     {
       "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": 28210.800931182683,
-      "required_pages_per_second_upper_estimate": 0.0016911944404255264
+      "required_token_positions_per_second": 28517.499630358652,
+      "required_pages_per_second_upper_estimate": 0.0017095805591393278
     },
     {
       "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": 28210.800931182683,
-      "required_pages_per_second_upper_estimate": 0.002193265686199991
+      "required_token_positions_per_second": 28517.499630358652,
+      "required_pages_per_second_upper_estimate": 0.0022171101610359146
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28210.800931182683,
-      "required_pages_per_second_upper_estimate": 0.36185786394713615
+      "required_token_positions_per_second": 28517.499630358652,
+      "required_pages_per_second_upper_estimate": 0.3657918655527591
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11427849948438507,
+    "producer_batch_seconds": 0.13305542778101653,
     "producer_p95_seconds": 0.2983669169989298,
-    "gpu_update_seconds": 0.9800501611933821,
-    "producer_batches_per_second": 8.750552418100654,
-    "gpu_batches_per_second": 1.020355936457707,
-    "producer_headroom": 8.575980307890685,
+    "gpu_update_seconds": 0.9695099625973865,
+    "producer_batches_per_second": 7.515664837407508,
+    "gpu_batches_per_second": 1.0314489160285971,
+    "producer_headroom": 7.286511935409446,
     "sustainable": true,
-    "buffer_seconds": 60.76310999398969,
+    "buffer_seconds": 60.10961768103797,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 236408.0604459258,
+    "generation_bytes_per_second": 236671.45770504605,
     "upload_bytes_per_second": 3750622.451129016,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.727666473706535,
+    "estimated_full_shards_per_hour": 50.78418539393936,
     "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": [

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -326,3 +326,4 @@
 {"step": 832600, "loss": 0.5355899333953857, "lr": 3e-05, "grad_norm": 0.2973308184618766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121869173641, "tokens_per_second": 138589.4877518306, "data_wait_s": 9.17899888008833e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 832700, "loss": 0.561130166053772, "lr": 3e-05, "grad_norm": 0.2903270618468921, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121884005720, "tokens_per_second": 153546.51378122595, "data_wait_s": 2.0450999727472663e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 832800, "loss": 0.5387684106826782, "lr": 3e-05, "grad_norm": 0.2740487478532095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121898765655, "tokens_per_second": 153421.251204086, "data_wait_s": 1.0067997209262103e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 832900, "loss": 0.524799108505249, "lr": 3e-05, "grad_norm": 0.2792230709649194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121913337115, "tokens_per_second": 150326.6721979659, "data_wait_s": 7.94499646872282e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3af69f5cabdca208c1e6f0d9046572d479c35fb0db6b71d1efc4affad633be3e
-size 699567
+oid sha256:fa4bedafe674438a69dbba0b2231eff673f39592afc30e52fb853981191dd109
+size 701685