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": 75870.51163640156,
-      "required_pages_per_second_upper_estimate": 0.005248898038847364
+      "required_token_positions_per_second": 73917.83828699337,
+      "required_pages_per_second_upper_estimate": 0.005113807565708909
     },
     {
       "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": 9483.813954550194,
-      "required_pages_per_second_upper_estimate": 0.003970674858214398
+      "required_token_positions_per_second": 9239.729785874171,
+      "required_pages_per_second_upper_estimate": 0.003868481913846795
     },
     {
       "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": 18967.62790910039,
-      "required_pages_per_second_upper_estimate": 0.003950456683479983
+      "required_token_positions_per_second": 18479.459571748343,
+      "required_pages_per_second_upper_estimate": 0.003848784092674338
     },
     {
       "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": 28451.441863650583,
-      "required_pages_per_second_upper_estimate": 0.0017056204968895485
+      "required_token_positions_per_second": 27719.189357622512,
+      "required_pages_per_second_upper_estimate": 0.001661723077238003
     },
     {
       "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": 28451.441863650583,
-      "required_pages_per_second_upper_estimate": 0.002211974460231771
+      "required_token_positions_per_second": 27719.189357622512,
+      "required_pages_per_second_upper_estimate": 0.0021550450487264738
     },
     {
       "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": 28451.441863650583,
-      "required_pages_per_second_upper_estimate": 0.36494454744873184
+      "required_token_positions_per_second": 27719.189357622512,
+      "required_pages_per_second_upper_estimate": 0.3555519985328884
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12499702303134086,
-    "producer_p95_seconds": 0.26965382200069143,
-    "gpu_update_seconds": 0.9717609438740938,
-    "producer_batches_per_second": 8.000190530531812,
-    "gpu_batches_per_second": 1.029059673887825,
-    "producer_headroom": 7.7742727011221815,
+    "producer_batch_seconds": 0.26419823575031387,
+    "producer_p95_seconds": 0.2708131470026274,
+    "gpu_update_seconds": 0.9974317662503022,
+    "producer_batches_per_second": 3.785036630392457,
+    "gpu_batches_per_second": 1.0025748465575273,
+    "producer_headroom": 3.7753157715744408,
     "sustainable": true,
-    "buffer_seconds": 60.249178520193816,
+    "buffer_seconds": 60.843337741268435,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389893.77725175617,
-    "upload_bytes_per_second": 5129626.9773589,
+    "generation_bytes_per_second": 381626.68331346125,
+    "upload_bytes_per_second": 5143598.4873222755,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.66212833561434,
+    "estimated_full_shards_per_hour": 81.88820242455367,
     "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": 1.6056268569991516,
-      "bytes": 7578967,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.20693444599965,
       "bytes": 6453330,
@@ -562,6 +556,12 @@
       "bytes": 11268683,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.0690768630011007,
+      "bytes": 5982757,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -206,3 +206,4 @@
 {"step": 740600, "loss": 0.49667567014694214, "lr": 5e-05, "grad_norm": 0.24028139674629126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108404032609, "tokens_per_second": 149848.78378891034, "data_wait_s": 8.445000275969505e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 740700, "loss": 0.49916335940361023, "lr": 5e-05, "grad_norm": 0.23138529878989852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108418555112, "tokens_per_second": 149006.88255348176, "data_wait_s": 1.749199873302132e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 740800, "loss": 0.5205853581428528, "lr": 5e-05, "grad_norm": 0.2212010544984674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108433043589, "tokens_per_second": 147755.80598715667, "data_wait_s": 1.9914001313736662e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 740900, "loss": 0.5532585382461548, "lr": 5e-05, "grad_norm": 0.22479074977986213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108447557034, "tokens_per_second": 149233.26902220756, "data_wait_s": 1.9133000023430213e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:98447ced5aaf53ca070622eb9a496c4551017298a2536043c7836f4fa683edfc
-size 443841
+oid sha256:0b8e425d8a88d95b6a30c8301ea775bdd0bef3978b1b6cfc60632ba34105ded3
+size 445959