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": 75414.82490767069,
-      "required_pages_per_second_upper_estimate": 0.006536765885870627
+      "required_token_positions_per_second": 75116.92299898001,
+      "required_pages_per_second_upper_estimate": 0.006510944503450804
     },
     {
       "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": 9426.853113458836,
-      "required_pages_per_second_upper_estimate": 0.003760512650972888
+      "required_token_positions_per_second": 9389.615374872501,
+      "required_pages_per_second_upper_estimate": 0.0037456579602969924
     },
     {
       "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": 18853.70622691767,
-      "required_pages_per_second_upper_estimate": 0.004361473121700661
+      "required_token_positions_per_second": 18779.230749745002,
+      "required_pages_per_second_upper_estimate": 0.004344244530780396
     },
     {
       "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": 28280.55934037651,
-      "required_pages_per_second_upper_estimate": 0.0017221376872012205
+      "required_token_positions_per_second": 28168.846124617503,
+      "required_pages_per_second_upper_estimate": 0.0017153349384754428
     },
     {
       "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": 28280.55934037651,
-      "required_pages_per_second_upper_estimate": 0.002177115743620939
+      "required_token_positions_per_second": 28168.846124617503,
+      "required_pages_per_second_upper_estimate": 0.002168515751029837
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28280.55934037651,
-      "required_pages_per_second_upper_estimate": 0.3511498980639521
+      "required_token_positions_per_second": 28168.846124617503,
+      "required_pages_per_second_upper_estimate": 0.34976279380430736
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1152253453594767,
-    "producer_p95_seconds": 0.4034826810020604,
-    "gpu_update_seconds": 0.9776327146587446,
-    "producer_batches_per_second": 8.678646150985522,
-    "gpu_batches_per_second": 1.0228790270680161,
-    "producer_headroom": 8.48452839615064,
+    "producer_batch_seconds": 0.1121480249219644,
+    "producer_p95_seconds": 0.37922541899752105,
+    "gpu_update_seconds": 0.9815098523271665,
+    "producer_batches_per_second": 8.916786547920275,
+    "gpu_batches_per_second": 1.018838473835992,
+    "producer_headroom": 8.751913847882093,
     "sustainable": true,
-    "buffer_seconds": 60.61322830884216,
+    "buffer_seconds": 60.85361084428432,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346780.9633111328,
-    "upload_bytes_per_second": 3718135.8829697864,
+    "generation_bytes_per_second": 359549.5091496398,
+    "upload_bytes_per_second": 3679521.780789209,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.41112207890022,
+    "estimated_full_shards_per_hour": 77.15095477930923,
     "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": 0.9655926410050597,
-      "bytes": 5648372,
-      "pages": 25,
-      "shards": 2
-    },
     {
       "seconds": 2.317049978002615,
       "bytes": 7094864,
@@ -562,6 +556,12 @@
       "bytes": 10876587,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 0.8632450160002918,
+      "bytes": 820169,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -398,3 +398,4 @@
 {"step": 39800, "loss": 0.6376858409494162, "lr": 0.0003, "grad_norm": 0.12891983290411593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5824231971, "tokens_per_second": 150190.43901381618, "data_wait_s": 9.909002983476967e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 39900, "loss": 0.6341990288347006, "lr": 0.0003, "grad_norm": 0.13038650884972652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5838941211, "tokens_per_second": 138312.6615532331, "data_wait_s": 1.079600042430684e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 40000, "loss": 0.659086462110281, "lr": 0.0003, "grad_norm": 0.13768149767346285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5853513827, "tokens_per_second": 148643.90850944628, "data_wait_s": 1.470099959988147e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 40100, "loss": 0.6127707604318857, "lr": 0.0003, "grad_norm": 0.14291243084088415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5868007171, "tokens_per_second": 134005.07927723238, "data_wait_s": 1.1392003216315061e-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:b4db0457abc340dfee14d69a99223e2736a148670a4cad94355de4dc6c8f8303
-size 847855
+oid sha256:05c1bfcf188b2545ceb27d873df1ec3ac1541e0c88ccbe60dd227a3025a88632
+size 849973