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": 77145.51281026368,
-      "required_pages_per_second_upper_estimate": 0.005337105578465459
+      "required_token_positions_per_second": 76296.27332002293,
+      "required_pages_per_second_upper_estimate": 0.005278353220023503
     },
     {
       "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": 9643.18910128296,
-      "required_pages_per_second_upper_estimate": 0.004037401904019889
+      "required_token_positions_per_second": 9537.034165002866,
+      "required_pages_per_second_upper_estimate": 0.00399295704896656
     },
     {
       "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": 19286.37820256592,
-      "required_pages_per_second_upper_estimate": 0.004016843963598335
+      "required_token_positions_per_second": 19074.06833000573,
+      "required_pages_per_second_upper_estimate": 0.003972625416131903
     },
     {
       "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": 28929.567303848882,
-      "required_pages_per_second_upper_estimate": 0.001734283386974161
+      "required_token_positions_per_second": 28611.1024950086,
+      "required_pages_per_second_upper_estimate": 0.0017151919079517932
     },
     {
       "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": 28929.567303848882,
-      "required_pages_per_second_upper_estimate": 0.0022491466101556344
+      "required_token_positions_per_second": 28611.1024950086,
+      "required_pages_per_second_upper_estimate": 0.0022243873720469575
     },
     {
       "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": 28929.567303848882,
-      "required_pages_per_second_upper_estimate": 0.3710774272244953
+      "required_token_positions_per_second": 28611.1024950086,
+      "required_pages_per_second_upper_estimate": 0.3669925025975629
     }
   ],
   "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": 0.3951263455080607,
-    "producer_p95_seconds": 0.2589216199994553,
-    "gpu_update_seconds": 0.9557004330417905,
-    "producer_batches_per_second": 2.5308360512235186,
-    "gpu_batches_per_second": 1.0463529840801824,
-    "producer_headroom": 2.418721110112092,
+    "producer_batch_seconds": 0.09164795255477998,
+    "producer_p95_seconds": 0.24429477800003951,
+    "gpu_update_seconds": 0.966338155085893,
+    "producer_batches_per_second": 10.911318497838543,
+    "gpu_batches_per_second": 1.0348344363067346,
+    "producer_headroom": 10.544023386755876,
     "sustainable": true,
-    "buffer_seconds": 60.2091272816328,
+    "buffer_seconds": 60.87930377041126,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 259219.26221638508,
-    "upload_bytes_per_second": 5422540.174717165,
+    "generation_bytes_per_second": 259159.63099174644,
+    "upload_bytes_per_second": 5403444.847318753,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.62241935604729,
+    "estimated_full_shards_per_hour": 55.609623883383705,
     "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": 1.2270069759997568,
-      "bytes": 11800938,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0093337090002024,
       "bytes": 5575178,
@@ -562,6 +556,12 @@
       "bytes": 761598,
       "pages": 23,
       "shards": 1
+    },
+    {
+      "seconds": 1.654032960999757,
+      "bytes": 12595480,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -240,3 +240,4 @@
 {"step": 709000, "loss": 0.5492202639579773, "lr": 8e-05, "grad_norm": 0.2270388262767895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103777427626, "tokens_per_second": 152381.12924728592, "data_wait_s": 1.1015999916708097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 709100, "loss": 0.5522074103355408, "lr": 8e-05, "grad_norm": 0.21061085954004322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103792103834, "tokens_per_second": 118702.04479120397, "data_wait_s": 1.6681999113643542e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 709200, "loss": 0.5702271461486816, "lr": 8e-05, "grad_norm": 0.23457596407394352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103806820972, "tokens_per_second": 152595.79665767297, "data_wait_s": 1.3566997949965298e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 709300, "loss": 0.5885176062583923, "lr": 8e-05, "grad_norm": 0.2977584378423985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103821490453, "tokens_per_second": 153195.24545300388, "data_wait_s": 1.198000245494768e-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:bb78b81f3b09e220c15f355df0e3d938aebe44f0ac3edaab790c9a105903e8be
-size 515853
+oid sha256:eeb25ccf3ee953c82191ec47aa1464016de0a86852ed5bfb50a0cbdd0a9c4607
+size 517971