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": 77053.4450475881,
-      "required_pages_per_second_upper_estimate": 0.0053307361170169204
+      "required_token_positions_per_second": 77058.28095775927,
+      "required_pages_per_second_upper_estimate": 0.0053310706765034195
     },
     {
       "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": 9631.680630948513,
-      "required_pages_per_second_upper_estimate": 0.004032583547815044
+      "required_token_positions_per_second": 9632.285119719909,
+      "required_pages_per_second_upper_estimate": 0.0040328366346404675
     },
     {
       "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": 19263.361261897026,
-      "required_pages_per_second_upper_estimate": 0.004012050141854549
+      "required_token_positions_per_second": 19264.570239439818,
+      "required_pages_per_second_upper_estimate": 0.00401230193999383
     },
     {
       "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": 28895.041892845536,
-      "required_pages_per_second_upper_estimate": 0.0017322136413017604
+      "required_token_positions_per_second": 28896.855359159727,
+      "required_pages_per_second_upper_estimate": 0.0017323223558382934
     },
     {
       "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": 28895.041892845536,
-      "required_pages_per_second_upper_estimate": 0.002246462411311358
+      "required_token_positions_per_second": 28896.855359159727,
+      "required_pages_per_second_upper_estimate": 0.002246603400340691
     },
     {
       "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": 28895.041892845536,
-      "required_pages_per_second_upper_estimate": 0.3706345723226426
+      "required_token_positions_per_second": 28896.855359159727,
+      "required_pages_per_second_upper_estimate": 0.37065783352137255
     }
   ],
   "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": 0.044854045585935864,
-    "producer_p95_seconds": 0.14449451699988458,
-    "gpu_update_seconds": 0.9568423573334806,
-    "producer_batches_per_second": 22.294533011166184,
-    "gpu_batches_per_second": 1.045104235128962,
-    "producer_headroom": 21.332353522053353,
+    "producer_batch_seconds": 0.08906162496092307,
+    "producer_p95_seconds": 0.13772478200007754,
+    "gpu_update_seconds": 0.9567823092292337,
+    "producer_batches_per_second": 11.228180492313752,
+    "gpu_batches_per_second": 1.045169826358497,
+    "producer_headroom": 10.742924459878587,
     "sustainable": true,
-    "buffer_seconds": 60.28106851200928,
+    "buffer_seconds": 60.27728548144172,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 258235.99024876632,
+    "generation_bytes_per_second": 275145.4798154911,
     "upload_bytes_per_second": 4640194.686183509,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.41143208119623,
+    "estimated_full_shards_per_hour": 59.03981490944434,
     "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

@@ -13,3 +13,4 @@
 {"step": 605700, "loss": 0.5830405354499817, "lr": 0.0001, "grad_norm": 0.13374721570454232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88656130069, "tokens_per_second": 152483.22327389175, "data_wait_s": 1.1977999974988052e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 605800, "loss": 0.5844405889511108, "lr": 0.0001, "grad_norm": 0.14258260715490792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88670798435, "tokens_per_second": 120628.73178277216, "data_wait_s": 1.3781999996353989e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 605900, "loss": 0.5470496416091919, "lr": 0.0001, "grad_norm": 0.14611003501020484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88685580969, "tokens_per_second": 153114.47393994956, "data_wait_s": 1.0173999953622115e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 606000, "loss": 0.5668455362319946, "lr": 0.0001, "grad_norm": 0.14701717059723388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88700396374, "tokens_per_second": 153547.1459799396, "data_wait_s": 1.07800001387659e-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:bb0411f08571135565a543969b24df9dc9d9ea31e0dcead93d679504dc17e487
-size 32718
+oid sha256:caff9ad657d4d2cc3d940981df95dbf03dbe63c30ba70815cc5a094bf7e0c960
+size 34836