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": 76410.49206095206,
-      "required_pages_per_second_upper_estimate": 0.005286255137544973
+      "required_token_positions_per_second": 76179.22423346803,
+      "required_pages_per_second_upper_estimate": 0.005270255492624337
     },
     {
       "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": 9551.311507619008,
-      "required_pages_per_second_upper_estimate": 0.003998934674175122
+      "required_token_positions_per_second": 9522.403029183504,
+      "required_pages_per_second_upper_estimate": 0.0039868312979318525
     },
     {
       "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": 19102.623015238016,
-      "required_pages_per_second_upper_estimate": 0.003978572604028099
+      "required_token_positions_per_second": 19044.80605836701,
+      "required_pages_per_second_upper_estimate": 0.003966530856647554
     },
     {
       "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": 28653.934522857024,
-      "required_pages_per_second_upper_estimate": 0.0017177596226205865
+      "required_token_positions_per_second": 28567.209087550513,
+      "required_pages_per_second_upper_estimate": 0.0017125605651960335
     },
     {
       "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": 28653.934522857024,
-      "required_pages_per_second_upper_estimate": 0.0022277173738139925
+      "required_token_positions_per_second": 28567.209087550513,
+      "required_pages_per_second_upper_estimate": 0.002220974852683787
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28653.934522857024,
-      "required_pages_per_second_upper_estimate": 0.36754190586135405
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28567.209087550513,
+      "required_pages_per_second_upper_estimate": 0.36642948509575957
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08254387310951472,
-    "producer_p95_seconds": 0.22626898999988043,
-    "gpu_update_seconds": 0.9648936685447299,
-    "producer_batches_per_second": 12.114769544109645,
-    "gpu_batches_per_second": 1.0363836271287987,
-    "producer_headroom": 11.68946442898992,
+    "producer_batch_seconds": 0.061474662273383274,
+    "producer_p95_seconds": 0.21363885999926424,
+    "gpu_update_seconds": 0.967822929963743,
+    "producer_batches_per_second": 16.266864477480354,
+    "gpu_batches_per_second": 1.033246856465224,
+    "producer_headroom": 15.743444439918166,
     "sustainable": true,
-    "buffer_seconds": 60.78830111831798,
+    "buffer_seconds": 60.00502165775207,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 529541.4258084036,
-    "upload_bytes_per_second": 5454454.944979651,
+    "generation_bytes_per_second": 512890.9300358155,
+    "upload_bytes_per_second": 5431235.839580323,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.62726288498956,
+    "estimated_full_shards_per_hour": 110.05445409589623,
     "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": 1.270762184999512,
-      "bytes": 12213784,
-      "pages": 48,
-      "shards": 4
-    },
     {
       "seconds": 1.305295600000136,
       "bytes": 1053111,
@@ -562,6 +556,12 @@
       "bytes": 1839915,
       "pages": 12,
       "shards": 2
+    },
+    {
+      "seconds": 1.339241851999759,
+      "bytes": 10779600,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -150,3 +150,4 @@
 {"step": 775000, "loss": 0.49935993552207947, "lr": 3e-05, "grad_norm": 0.2471969544188181, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113437606124, "tokens_per_second": 150626.16408695595, "data_wait_s": 1.0480000128154643e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 775100, "loss": 0.5219948291778564, "lr": 3e-05, "grad_norm": 0.2515800013685113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113452290848, "tokens_per_second": 141605.58837923338, "data_wait_s": 1.4733999705640599e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 775200, "loss": 0.5444227457046509, "lr": 3e-05, "grad_norm": 0.25118968262756286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113466870528, "tokens_per_second": 151271.90771167693, "data_wait_s": 1.2540000170702115e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 775300, "loss": 0.5524370670318604, "lr": 3e-05, "grad_norm": 0.25227665510208314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113481562555, "tokens_per_second": 151825.27060403733, "data_wait_s": 1.9462000636849552e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f48ef3360bdbd8243b4582470991e6ae57f02dd39e3a198de34919785d926eec
-size 324450
+oid sha256:a488a45e12cbbfb9b138aed2995428af3f001b251b4f8ede615a13cc2df50614
+size 326568