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": 76947.39081445568,
-      "required_pages_per_second_upper_estimate": 0.005323399039089095
+      "required_token_positions_per_second": 76847.29649176687,
+      "required_pages_per_second_upper_estimate": 0.005316474281594657
     },
     {
       "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": 9618.42385180696,
-      "required_pages_per_second_upper_estimate": 0.004027033211221505
+      "required_token_positions_per_second": 9605.912061470859,
+      "required_pages_per_second_upper_estimate": 0.004021794785883672
     },
     {
       "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": 19236.84770361392,
-      "required_pages_per_second_upper_estimate": 0.004006528066873729
+      "required_token_positions_per_second": 19211.824122941718,
+      "required_pages_per_second_upper_estimate": 0.004001316314935909
     },
     {
       "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": 28855.27155542088,
-      "required_pages_per_second_upper_estimate": 0.0017298294702989415
+      "required_token_positions_per_second": 28817.736184412577,
+      "required_pages_per_second_upper_estimate": 0.0017275792821201867
     },
     {
       "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": 28855.27155542088,
-      "required_pages_per_second_upper_estimate": 0.0022433704425078183
+      "required_token_positions_per_second": 28817.736184412577,
+      "required_pages_per_second_upper_estimate": 0.0022404522325125712
     },
     {
       "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": 28855.27155542088,
-      "required_pages_per_second_upper_estimate": 0.37012444113622045
+      "required_token_positions_per_second": 28817.736184412577,
+      "required_pages_per_second_upper_estimate": 0.3696429776992673
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5102352225859477,
-    "producer_p95_seconds": 0.3384576399985235,
-    "gpu_update_seconds": 0.9581611438623222,
-    "producer_batches_per_second": 1.9598803762152128,
-    "gpu_batches_per_second": 1.043665782531137,
-    "producer_headroom": 1.8778812231076867,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.3641520633263,
+    "buffer_seconds": 60.442776936175406,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311962.8383140146,
+    "generation_bytes_per_second": 312241.0344079271,
     "upload_bytes_per_second": 3770300.7460472737,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.93996297898606,
+    "estimated_full_shards_per_hour": 66.99965738466605,
     "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": [

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

@@ -402,3 +402,4 @@
 {"step": 882700, "loss": 0.5182936787605286, "lr": 3e-05, "grad_norm": 0.3073328887564094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129202265162, "tokens_per_second": 151357.08025770652, "data_wait_s": 1.7841004591900855e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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:d9bf3ff0c998de29ba08d83bd2869bf4c1ca5e7cc84d16702e32ed39ac9e5e05
-size 862101
+oid sha256:218ad775d54d1b5370f61815828581bd6015a37941881d38ca8ce84e8225f338
+size 864219