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": 76767.02941554454,
-      "required_pages_per_second_upper_estimate": 0.00531092121901632
+      "required_token_positions_per_second": 77446.61056526391,
+      "required_pages_per_second_upper_estimate": 0.005357936219799421
     },
     {
       "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": 9595.878676943068,
-      "required_pages_per_second_upper_estimate": 0.00401759401730278
+      "required_token_positions_per_second": 9680.826320657989,
+      "required_pages_per_second_upper_estimate": 0.004053159821817699
     },
     {
       "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": 19191.757353886136,
-      "required_pages_per_second_upper_estimate": 0.003997136936137919
+      "required_token_positions_per_second": 19361.652641315977,
+      "required_pages_per_second_upper_estimate": 0.0040325216440695285
     },
     {
       "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": 28787.636030829202,
-      "required_pages_per_second_upper_estimate": 0.0017257748238731366
+      "required_token_positions_per_second": 29042.478961973964,
+      "required_pages_per_second_upper_estimate": 0.0017410522684726403
     },
     {
       "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": 28787.636030829202,
-      "required_pages_per_second_upper_estimate": 0.0022381120779680474
+      "required_token_positions_per_second": 29042.478961973964,
+      "required_pages_per_second_upper_estimate": 0.0022579249949289627
     },
     {
       "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": 28787.636030829202,
-      "required_pages_per_second_upper_estimate": 0.36925688524812667
+      "required_token_positions_per_second": 29042.478961973964,
+      "required_pages_per_second_upper_estimate": 0.37252573673983097
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14662140923414313,
-    "producer_p95_seconds": 0.31913288599753287,
-    "gpu_update_seconds": 0.9604123093119301,
-    "producer_batches_per_second": 6.820286377162539,
-    "gpu_batches_per_second": 1.0412194744946905,
-    "producer_headroom": 6.550286989659371,
+    "producer_batch_seconds": 0.1457336140075256,
+    "producer_p95_seconds": 0.34086073800062877,
+    "gpu_update_seconds": 0.9519848507491461,
+    "producer_batches_per_second": 6.8618349089206045,
+    "gpu_batches_per_second": 1.0504368837519518,
+    "producer_headroom": 6.532362881634062,
     "sustainable": true,
-    "buffer_seconds": 60.5059754866516,
+    "buffer_seconds": 60.92703044794535,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 267165.06890059286,
-    "upload_bytes_per_second": 4586579.560007765,
+    "generation_bytes_per_second": 275376.9305475841,
+    "upload_bytes_per_second": 4605951.054097956,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.327404501565354,
+    "estimated_full_shards_per_hour": 59.089478848654196,
     "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": 1.7241964509994432,
-      "bytes": 5999571,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.129534389001492,
       "bytes": 2663790,
@@ -562,6 +556,12 @@
       "bytes": 11634070,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.6460893040020892,
+      "bytes": 7447219,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -297,3 +297,4 @@
 {"step": 872200, "loss": 0.5043037533760071, "lr": 3e-05, "grad_norm": 0.30055156094455737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127665666258, "tokens_per_second": 149899.29615613303, "data_wait_s": 1.6519999917363748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 872300, "loss": 0.4845273494720459, "lr": 3e-05, "grad_norm": 0.31965501601472557, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127680161674, "tokens_per_second": 150525.67748754594, "data_wait_s": 1.1422998795751482e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 872400, "loss": 0.5315535068511963, "lr": 3e-05, "grad_norm": 0.3236989336393728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127694707962, "tokens_per_second": 151667.05180677318, "data_wait_s": 1.0137999197468162e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 872500, "loss": 0.5857591032981873, "lr": 3e-05, "grad_norm": 0.3249897394419504, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127709264712, "tokens_per_second": 112082.88596471876, "data_wait_s": 1.1063999409088865e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b36c33b1d80e5deb1c67b4f40ffe131ad88c9a6afab9c19f40569a5316d31c62
-size 638145
+oid sha256:12e9434076f1d8be684b5451cb8d17b30a0999f24e533c764ef0dff84ebcd1b5
+size 640263