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": 76179.22423346803,
-      "required_pages_per_second_upper_estimate": 0.005270255492624337
+      "required_token_positions_per_second": 76314.73599962704,
+      "required_pages_per_second_upper_estimate": 0.005279630511037834
     },
     {
       "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": 9522.403029183504,
-      "required_pages_per_second_upper_estimate": 0.0039868312979318525
+      "required_token_positions_per_second": 9539.34199995338,
+      "required_pages_per_second_upper_estimate": 0.003993923291267266
     },
     {
       "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": 19044.80605836701,
-      "required_pages_per_second_upper_estimate": 0.003966530856647554
+      "required_token_positions_per_second": 19078.68399990676,
+      "required_pages_per_second_upper_estimate": 0.0039735867384488865
     },
     {
       "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": 28567.209087550513,
-      "required_pages_per_second_upper_estimate": 0.0017125605651960335
+      "required_token_positions_per_second": 28618.02599986014,
+      "required_pages_per_second_upper_estimate": 0.0017156069614960633
     },
     {
       "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": 28567.209087550513,
-      "required_pages_per_second_upper_estimate": 0.002220974852683787
+      "required_token_positions_per_second": 28618.02599986014,
+      "required_pages_per_second_upper_estimate": 0.0022249256440958856
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28567.209087550513,
-      "required_pages_per_second_upper_estimate": 0.36642948509575957
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28618.02599986014,
+      "required_pages_per_second_upper_estimate": 0.3670813098839181
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "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,
+    "producer_batch_seconds": 0.21162653931249054,
+    "producer_p95_seconds": 0.24317574399901787,
+    "gpu_update_seconds": 0.9661043707254693,
+    "producer_batches_per_second": 4.725305262982101,
+    "gpu_batches_per_second": 1.035084852425497,
+    "producer_headroom": 4.5651380675790705,
     "sustainable": true,
-    "buffer_seconds": 60.00502165775207,
+    "buffer_seconds": 60.86457535570457,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 512890.9300358155,
-    "upload_bytes_per_second": 5431235.839580323,
+    "generation_bytes_per_second": 525862.6523307776,
+    "upload_bytes_per_second": 5229400.578746273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.05445409589623,
+    "estimated_full_shards_per_hour": 112.83788373415466,
     "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": 1.305295600000136,
-      "bytes": 1053111,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.13043969499995,
       "bytes": 7715241,
@@ -562,6 +556,12 @@
       "bytes": 10779600,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 4.318171066001014,
+      "bytes": 1094935,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -151,3 +151,4 @@
 {"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}
+{"step": 775400, "loss": 0.5622367858886719, "lr": 3e-05, "grad_norm": 0.2651238242944428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113496280085, "tokens_per_second": 152814.09469151936, "data_wait_s": 1.0277000910718925e-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:a488a45e12cbbfb9b138aed2995428af3f001b251b4f8ede615a13cc2df50614
-size 326568
+oid sha256:9da6351bcec44d2d05a1e5333d3ad7c28f2229382867e6ddbe23388680efe40a
+size 328686