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": 76326.20220512335,
-      "required_pages_per_second_upper_estimate": 0.0052804237697393315
+      "required_token_positions_per_second": 76983.91313346477,
+      "required_pages_per_second_upper_estimate": 0.0053259257378615995
     },
     {
       "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": 9540.775275640419,
-      "required_pages_per_second_upper_estimate": 0.0039945233738672295
+      "required_token_positions_per_second": 9622.989141683096,
+      "required_pages_per_second_upper_estimate": 0.004028944602758549
     },
     {
       "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": 19081.550551280838,
-      "required_pages_per_second_upper_estimate": 0.003974183765504063
+      "required_token_positions_per_second": 19245.978283366192,
+      "required_pages_per_second_upper_estimate": 0.004008429725846547
     },
     {
       "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": 28622.325826921257,
-      "required_pages_per_second_upper_estimate": 0.0017158647295629215
+      "required_token_positions_per_second": 28868.96742504929,
+      "required_pages_per_second_upper_estimate": 0.0017306505167707827
     },
     {
       "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": 28622.325826921257,
-      "required_pages_per_second_upper_estimate": 0.002225259936736947
+      "required_token_positions_per_second": 28868.96742504929,
+      "required_pages_per_second_upper_estimate": 0.002244435236129663
     },
     {
       "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": 28622.325826921257,
-      "required_pages_per_second_upper_estimate": 0.3671364634486635
+      "required_token_positions_per_second": 28868.96742504929,
+      "required_pages_per_second_upper_estimate": 0.3703001170463346
     }
   ],
   "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": 0.12881362895313941,
-    "producer_p95_seconds": 0.35838772900024196,
-    "gpu_update_seconds": 0.9659592364082155,
-    "producer_batches_per_second": 7.7631536983077,
-    "gpu_batches_per_second": 1.035240372790844,
-    "producer_headroom": 7.498890018536919,
+    "producer_batch_seconds": 0.17478102778085258,
+    "producer_p95_seconds": 0.3542577150001307,
+    "gpu_update_seconds": 0.9577065778947857,
+    "producer_batches_per_second": 5.721444785493766,
+    "gpu_batches_per_second": 1.0441611481861,
+    "producer_headroom": 5.4794653061292005,
     "sustainable": true,
-    "buffer_seconds": 60.85543189371757,
+    "buffer_seconds": 60.335514407371505,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306393.5686662387,
-    "upload_bytes_per_second": 4756906.093845341,
+    "generation_bytes_per_second": 314065.9051469638,
+    "upload_bytes_per_second": 4625280.783915119,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.74492735853549,
+    "estimated_full_shards_per_hour": 67.39123216444669,
     "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": 1.539863748002972,
-      "bytes": 11993569,
-      "pages": 43,
-      "shards": 3
-    },
     {
       "seconds": 1.4481774710002355,
       "bytes": 7324064,
@@ -562,6 +556,12 @@
       "bytes": 7658306,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 3.788112343994726,
+      "bytes": 10339888,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -436,3 +436,4 @@
 {"step": 688600, "loss": 0.5817644596099854, "lr": 0.0001, "grad_norm": 0.20325837312311798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100792624872, "tokens_per_second": 152159.29606921034, "data_wait_s": 1.4031000318937004e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688700, "loss": 0.5967218279838562, "lr": 0.0001, "grad_norm": 0.20523049831808476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100807133319, "tokens_per_second": 150947.85528205158, "data_wait_s": 1.0051997378468513e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688800, "loss": 0.5524390339851379, "lr": 0.0001, "grad_norm": 0.1908495949919119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100821681614, "tokens_per_second": 151533.3032512891, "data_wait_s": 1.2402000720612705e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 688900, "loss": 0.5299914479255676, "lr": 0.0001, "grad_norm": 0.18807557853217352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100836258126, "tokens_per_second": 151536.51777184106, "data_wait_s": 1.3773002137895674e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4fdea5016bc1d9a4e97e0f12dd6f25aa1969dfe6d292b7bd5e34834d30e85683
-size 934113
+oid sha256:7e5f828f3efa4ceba032e47ed744b7bccbceb54860d624fdb33f474bc697a81d
+size 936231