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": 77233.21564589404,
-      "required_pages_per_second_upper_estimate": 0.0053431730641329825
+      "required_token_positions_per_second": 75950.4112241774,
+      "required_pages_per_second_upper_estimate": 0.005254425677722283
     },
     {
       "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": 9654.151955736756,
-      "required_pages_per_second_upper_estimate": 0.00404199182224926
+      "required_token_positions_per_second": 9493.801403022175,
+      "required_pages_per_second_upper_estimate": 0.003974856394327976
     },
     {
       "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": 19308.30391147351,
-      "required_pages_per_second_upper_estimate": 0.004021410510543959
+      "required_token_positions_per_second": 18987.60280604435,
+      "required_pages_per_second_upper_estimate": 0.003954616927739954
     },
     {
       "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": 28962.455867210265,
-      "required_pages_per_second_upper_estimate": 0.0017362550061297472
+      "required_token_positions_per_second": 28481.40420906652,
+      "required_pages_per_second_upper_estimate": 0.0017074166937473814
     },
     {
       "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": 28962.455867210265,
-      "required_pages_per_second_upper_estimate": 0.00225170355129548
+      "required_token_positions_per_second": 28481.40420906652,
+      "required_pages_per_second_upper_estimate": 0.0022143039007974305
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28962.455867210265,
-      "required_pages_per_second_upper_estimate": 0.3714992864023071
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28481.40420906652,
+      "required_pages_per_second_upper_estimate": 0.3653288722446675
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0938644272968645,
-    "producer_p95_seconds": 0.255305724000209,
-    "gpu_update_seconds": 0.9546151792777207,
-    "producer_batches_per_second": 10.65366325452885,
-    "gpu_batches_per_second": 1.0475425299193528,
-    "producer_headroom": 10.170148657686521,
+    "producer_batch_seconds": 0.2823017335938971,
+    "producer_p95_seconds": 0.27576607700029854,
+    "gpu_update_seconds": 0.9707386544936846,
+    "producer_batches_per_second": 3.5423091005120853,
+    "gpu_batches_per_second": 1.0301433814043157,
+    "producer_headroom": 3.4386563700318358,
     "sustainable": true,
-    "buffer_seconds": 60.1407562944964,
+    "buffer_seconds": 60.185796578608446,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 407382.344058255,
+    "generation_bytes_per_second": 407098.71421255136,
     "upload_bytes_per_second": 5046186.268323344,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.41476765928972,
+    "estimated_full_shards_per_hour": 87.35390729696661,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -257,3 +257,4 @@
 {"step": 630100, "loss": 0.5305880904197693, "lr": 0.0001, "grad_norm": 0.16814634413983823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92230192318, "tokens_per_second": 131325.83251408767, "data_wait_s": 1.0489999112905934e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 630200, "loss": 0.5366817712783813, "lr": 0.0001, "grad_norm": 0.16661386707785586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92244652675, "tokens_per_second": 150567.7014132866, "data_wait_s": 1.4043998817214742e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 630300, "loss": 0.5650016665458679, "lr": 0.0001, "grad_norm": 0.1550158671266286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92259172622, "tokens_per_second": 150655.12316049854, "data_wait_s": 9.642997611081228e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 630400, "loss": 0.5558711290359497, "lr": 0.0001, "grad_norm": 0.18572500619261653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92273764615, "tokens_per_second": 151292.37098062615, "data_wait_s": 1.5177000022958964e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f404a01e429b7de5df7c5fe87b4cdf931bcc96a295122db227e8a8b66c7c6c16
-size 553425
+oid sha256:2ed268f26c9ca841741165e9b28258a9fdb65ca3d0aa1937619a357c36b20b6d
+size 555543