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": 77226.06750125453,
-      "required_pages_per_second_upper_estimate": 0.00534267853890084
+      "required_token_positions_per_second": 77380.8238906204,
+      "required_pages_per_second_upper_estimate": 0.005353384945001467
     },
     {
       "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": 9653.258437656816,
-      "required_pages_per_second_upper_estimate": 0.004041617724887968
+      "required_token_positions_per_second": 9672.60298632755,
+      "required_pages_per_second_upper_estimate": 0.004049716883456293
     },
     {
       "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": 19306.51687531363,
-      "required_pages_per_second_upper_estimate": 0.004021038318039169
+      "required_token_positions_per_second": 19345.2059726551,
+      "required_pages_per_second_upper_estimate": 0.004029096236715287
     },
     {
       "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": 28959.775312970443,
-      "required_pages_per_second_upper_estimate": 0.0017360943109960393
+      "required_token_positions_per_second": 29017.808958982652,
+      "required_pages_per_second_upper_estimate": 0.0017395733394622245
     },
     {
       "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": 28959.775312970443,
-      "required_pages_per_second_upper_estimate": 0.002251495149993846
+      "required_token_positions_per_second": 29017.808958982652,
+      "required_pages_per_second_upper_estimate": 0.0022560070107082632
     },
     {
       "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": 28959.775312970443,
-      "required_pages_per_second_upper_estimate": 0.3714649031306736
+      "required_token_positions_per_second": 29017.808958982652,
+      "required_pages_per_second_upper_estimate": 0.3722092964300439
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.19426115195307148,
-    "producer_p95_seconds": 0.25104810500124586,
-    "gpu_update_seconds": 0.9547035396927638,
-    "producer_batches_per_second": 5.147709616390901,
-    "gpu_batches_per_second": 1.0474455770026927,
-    "producer_headroom": 4.914536592078872,
+    "producer_batch_seconds": 0.11122698781241525,
+    "producer_p95_seconds": 0.2614132519993291,
+    "gpu_update_seconds": 0.9527941974902754,
+    "producer_batches_per_second": 8.990623765578404,
+    "gpu_batches_per_second": 1.049544594870611,
+    "producer_headroom": 8.566214155661273,
     "sustainable": true,
-    "buffer_seconds": 60.14632300064412,
+    "buffer_seconds": 60.02603444188735,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 485042.42964528635,
-    "upload_bytes_per_second": 3901781.291155653,
+    "generation_bytes_per_second": 485177.629263044,
+    "upload_bytes_per_second": 3861079.8789959704,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.07881419199889,
+    "estimated_full_shards_per_hour": 104.10782488268366,
     "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": 0.9967152519966476,
-      "bytes": 11168227,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0345216240020818,
       "bytes": 6241760,
@@ -562,6 +556,12 @@
       "bytes": 12031348,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 1.1874672249978175,
+      "bytes": 7293811,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -244,3 +244,4 @@
 {"step": 549400, "loss": 0.6239665150642395, "lr": 0.0003, "grad_norm": 0.18234078992583755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80414638023, "tokens_per_second": 152883.3989896266, "data_wait_s": 1.0659998224582523e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 549500, "loss": 0.6144766211509705, "lr": 0.0003, "grad_norm": 0.17061365900134132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80429278828, "tokens_per_second": 153154.0853361587, "data_wait_s": 7.97999746282585e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 549600, "loss": 0.5847856402397156, "lr": 0.0003, "grad_norm": 0.1970315840782169, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80443946447, "tokens_per_second": 153066.19631061135, "data_wait_s": 1.0759002179838717e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 549700, "loss": 0.5437107682228088, "lr": 0.0003, "grad_norm": 0.14214246667607797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80458549955, "tokens_per_second": 152045.07784303802, "data_wait_s": 1.4467997971223667e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:770cb0d119cd9b592bda194c365219a272ccd2e881b2b93c0d19c0e7301a5f1b
-size 524325
+oid sha256:1bc0fdd66c44b3e400b0867d2642eb17e433724e232f18750cfeb364d417251c
+size 526443