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": 76847.29649176687,
-      "required_pages_per_second_upper_estimate": 0.005316474281594657
+      "required_token_positions_per_second": 76234.7450452284,
+      "required_pages_per_second_upper_estimate": 0.005274096551207951
     },
     {
       "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": 9605.912061470859,
-      "required_pages_per_second_upper_estimate": 0.004021794785883672
+      "required_token_positions_per_second": 9529.34313065355,
+      "required_pages_per_second_upper_estimate": 0.0039897369734915615
     },
     {
       "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": 19211.824122941718,
-      "required_pages_per_second_upper_estimate": 0.004001316314935909
+      "required_token_positions_per_second": 19058.6862613071,
+      "required_pages_per_second_upper_estimate": 0.003969421736874409
     },
     {
       "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": 28817.736184412577,
-      "required_pages_per_second_upper_estimate": 0.0017275792821201867
+      "required_token_positions_per_second": 28588.02939196065,
+      "required_pages_per_second_upper_estimate": 0.001713808710654657
     },
     {
       "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": 28817.736184412577,
-      "required_pages_per_second_upper_estimate": 0.0022404522325125712
+      "required_token_positions_per_second": 28588.02939196065,
+      "required_pages_per_second_upper_estimate": 0.002222593539772834
     },
     {
       "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": 28817.736184412577,
-      "required_pages_per_second_upper_estimate": 0.3696429776992673
+      "required_token_positions_per_second": 28588.02939196065,
+      "required_pages_per_second_upper_estimate": 0.36669654560563164
     }
   ],
   "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": 0.1228444739998622,
-    "producer_p95_seconds": 0.3453624120011227,
-    "gpu_update_seconds": 0.9594091577170699,
-    "producer_batches_per_second": 8.140374307770015,
-    "gpu_batches_per_second": 1.0423081663922373,
-    "producer_headroom": 7.809949658119307,
+    "producer_batch_seconds": 0.1362329723360176,
+    "producer_p95_seconds": 0.32272735700098565,
+    "gpu_update_seconds": 0.9671180766231828,
+    "producer_batches_per_second": 7.34036689395213,
+    "gpu_batches_per_second": 1.033999905669873,
+    "producer_headroom": 7.0990015121874706,
     "sustainable": true,
-    "buffer_seconds": 60.442776936175406,
+    "buffer_seconds": 60.92843882726052,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312241.0344079271,
-    "upload_bytes_per_second": 3770300.7460472737,
+    "generation_bytes_per_second": 302578.2873911116,
+    "upload_bytes_per_second": 2983511.381412923,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.99965738466605,
+    "estimated_full_shards_per_hour": 64.92625681209574,
     "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": 1.4537843559992325,
-      "bytes": 4298944,
-      "pages": 20,
-      "shards": 2
-    },
     {
       "seconds": 1.740884680999443,
       "bytes": 961930,
@@ -562,6 +556,12 @@
       "bytes": 13222947,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 33.416039668998565,
+      "bytes": 11237253,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -403,3 +403,4 @@
 {"step": 882800, "loss": 0.5007718801498413, "lr": 3e-05, "grad_norm": 0.29883861408068824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129216915075, "tokens_per_second": 148817.6257671427, "data_wait_s": 1.154300116468221e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 882900, "loss": 0.5469985008239746, "lr": 3e-05, "grad_norm": 0.32545219418500004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129231586455, "tokens_per_second": 152132.23190774635, "data_wait_s": 6.903996109031141e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 883000, "loss": 0.5948290228843689, "lr": 3e-05, "grad_norm": 0.31355533686785125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129246235770, "tokens_per_second": 152058.33261111323, "data_wait_s": 1.2789001630153507e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 883100, "loss": 0.5442803502082825, "lr": 3e-05, "grad_norm": 0.32910081750747927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129260943821, "tokens_per_second": 153160.2571782582, "data_wait_s": 1.3328994100447744e-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 sha256:218ad775d54d1b5370f61815828581bd6015a37941881d38ca8ce84e8225f338
-size 864219
+oid sha256:5d36dc0a214d2d0c1b2ad59c59a17f5e99beb2b946b88ebbf1fcc17e36d7a80c
+size 866337