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": 77195.71180734529,
-      "required_pages_per_second_upper_estimate": 0.005340578461561285
+      "required_token_positions_per_second": 76850.63052901966,
+      "required_pages_per_second_upper_estimate": 0.005316704938027838
     },
     {
       "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": 9649.463975918161,
-      "required_pages_per_second_upper_estimate": 0.004040029062995365
+      "required_token_positions_per_second": 9606.328816127458,
+      "required_pages_per_second_upper_estimate": 0.004021969272355563
     },
     {
       "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": 19298.927951836322,
-      "required_pages_per_second_upper_estimate": 0.004019457745412215
+      "required_token_positions_per_second": 19212.657632254915,
+      "required_pages_per_second_upper_estimate": 0.00400148991294473
     },
     {
       "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": 28948.391927754485,
-      "required_pages_per_second_upper_estimate": 0.0017354118944337654
+      "required_token_positions_per_second": 28818.986448382377,
+      "required_pages_per_second_upper_estimate": 0.0017276542335361337
     },
     {
       "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": 28948.391927754485,
-      "required_pages_per_second_upper_estimate": 0.002250610141863525
+      "required_token_positions_per_second": 28818.986448382377,
+      "required_pages_per_second_upper_estimate": 0.0022405494350368933
     },
     {
       "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": 28948.391927754485,
-      "required_pages_per_second_upper_estimate": 0.37131888928765006
+      "required_token_positions_per_second": 28818.986448382377,
+      "required_pages_per_second_upper_estimate": 0.369659014743043
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.10110205546118323,
-    "producer_p95_seconds": 0.30273405700427247,
-    "gpu_update_seconds": 0.9550789580644125,
-    "producer_batches_per_second": 9.890995741267956,
-    "gpu_batches_per_second": 1.04703385155362,
-    "producer_headroom": 9.44668190678974,
+    "producer_batch_seconds": 0.08610134925771717,
+    "producer_p95_seconds": 0.30088820799574023,
+    "gpu_update_seconds": 0.9593675353406433,
+    "producer_batches_per_second": 11.61421985393999,
+    "gpu_batches_per_second": 1.042353387166608,
+    "producer_headroom": 11.142305476178775,
     "sustainable": true,
-    "buffer_seconds": 60.169974358057985,
+    "buffer_seconds": 60.44015472646053,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300360.95524304686,
-    "upload_bytes_per_second": 5548345.491218363,
+    "generation_bytes_per_second": 299424.887871059,
+    "upload_bytes_per_second": 5625310.026382966,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.45046894997172,
+    "estimated_full_shards_per_hour": 64.2496106824763,
     "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": 1.0567811459986842,
-      "bytes": 4557138,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.3797404860015376,
       "bytes": 7238899,
@@ -562,6 +556,12 @@
       "bytes": 2430044,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3565980959974695,
+      "bytes": 12199549,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -359,3 +359,4 @@
 {"step": 520900, "loss": 0.5985284447669983, "lr": 0.0003, "grad_norm": 0.14101838737853054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76242166162, "tokens_per_second": 154491.59539069392, "data_wait_s": 1.4563003787770867e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 521000, "loss": 0.6383951902389526, "lr": 0.0003, "grad_norm": 0.17458866481883503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76256789017, "tokens_per_second": 152092.41776261042, "data_wait_s": 1.4425997505895793e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 521100, "loss": 0.5398332476615906, "lr": 0.0003, "grad_norm": 0.16235187122789377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76271383304, "tokens_per_second": 152091.06721588864, "data_wait_s": 1.2454998795874417e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 521200, "loss": 0.6419379711151123, "lr": 0.0003, "grad_norm": 0.1925108938238185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76286146838, "tokens_per_second": 154367.94719962295, "data_wait_s": 9.009003406390548e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74c3d6e5ae1f60244ad75a0c78b98467455e32579b49f2a49f0841bf17bf6183
-size 770244
+oid sha256:5fba00244f187ec500469680931b5ded661937089cc26e77358b012bb89c1bbb
+size 772362