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": 75978.33044874344,
-      "required_pages_per_second_upper_estimate": 0.005256357194459287
+      "required_token_positions_per_second": 77437.87016246314,
+      "required_pages_per_second_upper_estimate": 0.005357331538453134
     },
     {
       "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": 9497.29130609293,
-      "required_pages_per_second_upper_estimate": 0.003976317543866238
+      "required_token_positions_per_second": 9679.733770307892,
+      "required_pages_per_second_upper_estimate": 0.004052702393801159
     },
     {
       "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": 18994.58261218586,
-      "required_pages_per_second_upper_estimate": 0.003956070637289364
+      "required_token_positions_per_second": 19359.467540615784,
+      "required_pages_per_second_upper_estimate": 0.004032066545218659
     },
     {
       "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": 28491.87391827879,
-      "required_pages_per_second_upper_estimate": 0.0017080443368283366
+      "required_token_positions_per_second": 29039.201310923676,
+      "required_pages_per_second_upper_estimate": 0.0017408557782968587
     },
     {
       "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": 28491.87391827879,
-      "required_pages_per_second_upper_estimate": 0.0022151178746373005
+      "required_token_positions_per_second": 29039.201310923676,
+      "required_pages_per_second_upper_estimate": 0.0022576701719766662
     },
     {
       "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": 28491.87391827879,
-      "required_pages_per_second_upper_estimate": 0.3654631664329446
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29039.201310923676,
+      "required_pages_per_second_upper_estimate": 0.3724836945514254
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.28005237713301767,
-    "producer_p95_seconds": 0.29375918800360523,
-    "gpu_update_seconds": 0.9703819439641207,
-    "producer_batches_per_second": 3.5707606206999833,
-    "gpu_batches_per_second": 1.0305220601229308,
-    "producer_headroom": 3.46500163254538,
+    "producer_batch_seconds": 0.10504363861713273,
+    "producer_p95_seconds": 0.29685203899862245,
+    "gpu_update_seconds": 0.9520923011611774,
+    "producer_batches_per_second": 9.519853016943179,
+    "gpu_batches_per_second": 1.0503183344518112,
+    "producer_headroom": 9.063778765617608,
     "sustainable": true,
-    "buffer_seconds": 60.16368052577548,
+    "buffer_seconds": 60.93390727431535,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383001.02526441426,
-    "upload_bytes_per_second": 4927176.7925582,
+    "generation_bytes_per_second": 399364.58575340774,
+    "upload_bytes_per_second": 4855161.255737214,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.18310421418497,
+    "estimated_full_shards_per_hour": 85.69434337093043,
     "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": 1.6013217930012615,
-      "bytes": 18143583,
-      "pages": 46,
-      "shards": 5
-    },
     {
       "seconds": 1.0924186460033525,
       "bytes": 6205338,
@@ -562,6 +556,12 @@
       "bytes": 6037077,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.49954066099599,
+      "bytes": 11232631,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -352,3 +352,4 @@
 {"step": 795200, "loss": 0.5257145762443542, "lr": 3e-05, "grad_norm": 0.2820711886502756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116394635879, "tokens_per_second": 151854.3283704057, "data_wait_s": 1.1088995961472392e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 795300, "loss": 0.5528821349143982, "lr": 3e-05, "grad_norm": 0.27490217585024596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116409220582, "tokens_per_second": 151332.2723023513, "data_wait_s": 8.536000677850097e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 795400, "loss": 0.47972771525382996, "lr": 3e-05, "grad_norm": 0.26239027678447313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116423830930, "tokens_per_second": 152853.67381406273, "data_wait_s": 1.5193996659945697e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 795500, "loss": 0.4830298125743866, "lr": 3e-05, "grad_norm": 0.26718392901315674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116438443906, "tokens_per_second": 152073.72058413352, "data_wait_s": 8.909999451134354e-06, "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 sha256:5ddebdcb8690f2fd79565ce7fe8286cdcf0275ff73536faac11d58ec9f6feffd
-size 755418
+oid sha256:927ead143b6b3d105260780810a268e64707d8f6aed02d2f55961b6df0443e48
+size 757536