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": 76438.21366744445,
-      "required_pages_per_second_upper_estimate": 0.005288172982605105
+      "required_token_positions_per_second": 76300.96337062248,
+      "required_pages_per_second_upper_estimate": 0.0052786776886063974
     },
     {
       "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": 9554.776708430556,
-      "required_pages_per_second_upper_estimate": 0.004000385481393295
+      "required_token_positions_per_second": 9537.62042132781,
+      "required_pages_per_second_upper_estimate": 0.00399320250224739
     },
     {
       "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": 19109.553416861112,
-      "required_pages_per_second_upper_estimate": 0.003980016023919208
+      "required_token_positions_per_second": 19075.24084265562,
+      "required_pages_per_second_upper_estimate": 0.003972869619595637
     },
     {
       "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": 28664.330125291668,
-      "required_pages_per_second_upper_estimate": 0.0017183828231133777
+      "required_token_positions_per_second": 28612.861263983432,
+      "required_pages_per_second_upper_estimate": 0.0017152973434663497
     },
     {
       "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": 28664.330125291668,
-      "required_pages_per_second_upper_estimate": 0.0022285255861778635
+      "required_token_positions_per_second": 28612.861263983432,
+      "required_pages_per_second_upper_estimate": 0.0022245241085987436
     },
     {
       "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": 28664.330125291668,
-      "required_pages_per_second_upper_estimate": 0.36767524948745745
+      "required_token_positions_per_second": 28612.861263983432,
+      "required_pages_per_second_upper_estimate": 0.36701506219755303
     }
   ],
   "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": 0.09179875850003327,
-    "producer_p95_seconds": 0.2641290530009428,
-    "gpu_update_seconds": 0.9645437335933094,
-    "producer_batches_per_second": 10.893393509233979,
-    "gpu_batches_per_second": 1.036759625480746,
-    "producer_headroom": 10.507154446897664,
+    "producer_batch_seconds": 0.08737870292983985,
+    "producer_p95_seconds": 0.2560379579990695,
+    "gpu_update_seconds": 0.9662787564276923,
+    "producer_batches_per_second": 11.444436303923434,
+    "gpu_batches_per_second": 1.0348980491892155,
+    "producer_headroom": 11.058515679771071,
     "sustainable": true,
-    "buffer_seconds": 60.76625521637849,
+    "buffer_seconds": 60.87556165494462,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405073.2925151763,
-    "upload_bytes_per_second": 5292571.983536604,
+    "generation_bytes_per_second": 370291.0517145556,
+    "upload_bytes_per_second": 5245571.346972623,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.91929895011393,
+    "estimated_full_shards_per_hour": 79.4558397634268,
     "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": 1.3652075969985162,
-      "bytes": 5700613,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.7661054199998034,
       "bytes": 7454533,
@@ -562,6 +556,12 @@
       "bytes": 6236357,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1350728899997193,
+      "bytes": 701024,
+      "pages": 21,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -250,3 +250,4 @@
 {"step": 745000, "loss": 0.5192437767982483, "lr": 5e-05, "grad_norm": 0.23177553583662464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109047827948, "tokens_per_second": 151153.61984531832, "data_wait_s": 1.3546999980462715e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 745100, "loss": 0.5198097229003906, "lr": 5e-05, "grad_norm": 0.22867919331413317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109062409782, "tokens_per_second": 135812.38742473102, "data_wait_s": 8.88399881660007e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 745200, "loss": 0.5324106812477112, "lr": 5e-05, "grad_norm": 0.24198127499305994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109076959884, "tokens_per_second": 149831.06108983708, "data_wait_s": 1.5336001524701715e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 745300, "loss": 0.5240442156791687, "lr": 5e-05, "grad_norm": 0.23911306087253575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109091459873, "tokens_per_second": 149986.14113557548, "data_wait_s": 1.0012001439463347e-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:d7574816e26a5ceb5a1931dded09c3100727c262bcefcbead87f409949398535
-size 537816
+oid sha256:b68b30f7f6b77b9309b74d0cefcef421736d36e5d0ec286bb55e1d52e22ca484
+size 539934