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": 76279.88003806896,
-      "required_pages_per_second_upper_estimate": 0.0052772190947402734
+      "required_token_positions_per_second": 76658.57767626128,
+      "required_pages_per_second_upper_estimate": 0.0053034182760499995
     },
     {
       "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": 9534.98500475862,
-      "required_pages_per_second_upper_estimate": 0.003992099108363626
+      "required_token_positions_per_second": 9582.32220953266,
+      "required_pages_per_second_upper_estimate": 0.004011918207489274
     },
     {
       "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": 19069.97000951724,
-      "required_pages_per_second_upper_estimate": 0.003971771844054129
+      "required_token_positions_per_second": 19164.64441906532,
+      "required_pages_per_second_upper_estimate": 0.003991490026830917
     },
     {
       "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": 28604.95501427586,
-      "required_pages_per_second_upper_estimate": 0.001714823375868525
+      "required_token_positions_per_second": 28746.96662859798,
+      "required_pages_per_second_upper_estimate": 0.0017233367553079566
     },
     {
       "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": 28604.95501427586,
-      "required_pages_per_second_upper_estimate": 0.0022239094324598974
+      "required_token_positions_per_second": 28746.96662859798,
+      "required_pages_per_second_upper_estimate": 0.002234950210830365
     },
     {
       "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": 28604.95501427586,
-      "required_pages_per_second_upper_estimate": 0.36691364931537385
+      "required_token_positions_per_second": 28746.96662859798,
+      "required_pages_per_second_upper_estimate": 0.3687352218237065
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0793257537345653,
-    "producer_p95_seconds": 0.27102510900294874,
-    "gpu_update_seconds": 0.9665458304759342,
-    "producer_batches_per_second": 12.606246432225975,
-    "gpu_batches_per_second": 1.0346120881899543,
-    "producer_headroom": 12.184514927020137,
+    "producer_batch_seconds": 0.0986103749687004,
+    "producer_p95_seconds": 0.28127252600097563,
+    "gpu_update_seconds": 0.961771040305007,
+    "producer_batches_per_second": 10.140920773472434,
+    "gpu_batches_per_second": 1.0397485036385266,
+    "producer_headroom": 9.75324392195324,
     "sustainable": true,
-    "buffer_seconds": 60.892387319983854,
+    "buffer_seconds": 60.59157553921544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 171,
+    "upload_pages_per_commit": 172,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 702722.5693498176,
-    "upload_bytes_per_second": 5134699.5794174895,
+    "generation_bytes_per_second": 714164.14453024,
+    "upload_bytes_per_second": 5123755.694367059,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.39395241914222,
+    "estimated_full_shards_per_hour": 76.62150026288224,
     "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": 1.2763601959995867,
-      "bytes": 5814992,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.203305928996997,
       "bytes": 7121753,
@@ -562,6 +556,12 @@
       "bytes": 5788082,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.5139052369995625,
+      "bytes": 6128565,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -317,3 +317,4 @@
 {"step": 751700, "loss": 0.5572493076324463, "lr": 5e-05, "grad_norm": 0.24770711443238044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110027801931, "tokens_per_second": 151018.90590762708, "data_wait_s": 1.3755001418758184e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 751800, "loss": 0.5707524418830872, "lr": 5e-05, "grad_norm": 0.25372712582124146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110042453269, "tokens_per_second": 151336.983288377, "data_wait_s": 1.3872999261366203e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 751900, "loss": 0.4962722361087799, "lr": 5e-05, "grad_norm": 0.24667855449304538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110056944555, "tokens_per_second": 149582.60326599845, "data_wait_s": 1.2260999938007444e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 752000, "loss": 0.5134796500205994, "lr": 5e-05, "grad_norm": 0.2307066487088763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110071415461, "tokens_per_second": 149253.18554392314, "data_wait_s": 1.3985001714900136e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0b84fef6b4b1178c363dd7244ed23eed7e1553d6942eb55e41474d9500e6376
-size 680505
+oid sha256:6a5536f83849330eab103b8b347da28f3529b0306b80e09c94e02532a4c96219
+size 682623