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": 77334.41655304712,
-      "required_pages_per_second_upper_estimate": 0.005350174377708288
+      "required_token_positions_per_second": 77100.11616749498,
+      "required_pages_per_second_upper_estimate": 0.005333964933383997
     },
     {
       "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": 9666.80206913089,
-      "required_pages_per_second_upper_estimate": 0.0040472881605629766
+      "required_token_positions_per_second": 9637.514520936873,
+      "required_pages_per_second_upper_estimate": 0.004035026075727695
     },
     {
       "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": 19333.60413826178,
-      "required_pages_per_second_upper_estimate": 0.004026679880572107
+      "required_token_positions_per_second": 19275.029041873746,
+      "required_pages_per_second_upper_estimate": 0.004014480232723649
     },
     {
       "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": 29000.406207392673,
-      "required_pages_per_second_upper_estimate": 0.0017385300710768684
+      "required_token_positions_per_second": 28912.54356281062,
+      "required_pages_per_second_upper_estimate": 0.0017332628396926645
     },
     {
       "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": 29000.406207392673,
-      "required_pages_per_second_upper_estimate": 0.002254654023318757
+      "required_token_positions_per_second": 28912.54356281062,
+      "required_pages_per_second_upper_estimate": 0.002247823089169433
     },
     {
       "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": 29000.406207392673,
-      "required_pages_per_second_upper_estimate": 0.3719860726182665
+      "required_token_positions_per_second": 28912.54356281062,
+      "required_pages_per_second_upper_estimate": 0.37085906495312554
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.050083175289056925,
-    "producer_p95_seconds": 0.13846570800001246,
-    "gpu_update_seconds": 0.9533659564034686,
-    "producer_batches_per_second": 19.96678513749303,
-    "gpu_batches_per_second": 1.0489151550706262,
-    "producer_headroom": 19.035653208908602,
+    "producer_batch_seconds": 0.06462485705466392,
+    "producer_p95_seconds": 0.1415127699997356,
+    "gpu_update_seconds": 0.9562631506265271,
+    "producer_batches_per_second": 15.473922041392443,
+    "gpu_batches_per_second": 1.045737252705824,
+    "producer_headroom": 14.797141443851201,
     "sustainable": true,
-    "buffer_seconds": 60.06205525341852,
+    "buffer_seconds": 60.24457848947121,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 71,
+    "upload_pages_per_commit": 72,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 275780.0652261353,
-    "upload_bytes_per_second": 4767625.9400929455,
+    "generation_bytes_per_second": 290750.9027435973,
+    "upload_bytes_per_second": 4629087.281196482,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.17598216617627,
+    "estimated_full_shards_per_hour": 62.38837539416255,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 7378325,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.7817219419998764,
+      "bytes": 1157795,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -15,3 +15,4 @@
 {"step": 605900, "loss": 0.5470496416091919, "lr": 0.0001, "grad_norm": 0.14611003501020484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88685580969, "tokens_per_second": 153114.47393994956, "data_wait_s": 1.0173999953622115e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 606000, "loss": 0.5668455362319946, "lr": 0.0001, "grad_norm": 0.14701717059723388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88700396374, "tokens_per_second": 153547.1459799396, "data_wait_s": 1.07800001387659e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 606100, "loss": 0.5854063630104065, "lr": 0.0001, "grad_norm": 0.13425116211854862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88715194408, "tokens_per_second": 153708.0282001785, "data_wait_s": 8.715000149095431e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 606200, "loss": 0.5686312913894653, "lr": 0.0001, "grad_norm": 0.14727125806159375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88729746436, "tokens_per_second": 150734.31198214568, "data_wait_s": 1.3697999747819267e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:45ffd6d01531f397bf0dae0b6917e73d3d904a85700bec1f67b75d718d2079c8
-size 36954
+oid sha256:584f3929ee39483837c9d4b3f49c595cd4ee256ec29ab5357183159bfddf03b1
+size 39072