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": 77491.67183817635,
-      "required_pages_per_second_upper_estimate": 0.005361053663216061
+      "required_token_positions_per_second": 77580.81305464669,
+      "required_pages_per_second_upper_estimate": 0.005367220659407598
     },
     {
       "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": 9686.458979772044,
-      "required_pages_per_second_upper_estimate": 0.004055518098565456
+      "required_token_positions_per_second": 9697.601631830836,
+      "required_pages_per_second_upper_estimate": 0.004060183294297438
     },
     {
       "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": 19372.917959544087,
-      "required_pages_per_second_upper_estimate": 0.004034867912770024
+      "required_token_positions_per_second": 19395.20326366167,
+      "required_pages_per_second_upper_estimate": 0.004039509353914726
     },
     {
       "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": 29059.376939316127,
-      "required_pages_per_second_upper_estimate": 0.0017420652764125854
+      "required_token_positions_per_second": 29092.804895492507,
+      "required_pages_per_second_upper_estimate": 0.0017440692313438258
     },
     {
       "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": 29059.376939316127,
-      "required_pages_per_second_upper_estimate": 0.0022592387383409684
+      "required_token_positions_per_second": 29092.804895492507,
+      "required_pages_per_second_upper_estimate": 0.0022618376148996417
     },
     {
       "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": 29059.376939316127,
-      "required_pages_per_second_upper_estimate": 0.3727424858495418
+      "required_token_positions_per_second": 29092.804895492507,
+      "required_pages_per_second_upper_estimate": 0.37317126377922943
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12681687072682735,
+    "producer_batch_seconds": 0.10908783268746447,
     "producer_p95_seconds": 0.3095279019980808,
-    "gpu_update_seconds": 0.9514312732078369,
-    "producer_batches_per_second": 7.8853861814180215,
-    "gpu_batches_per_second": 1.0510480663815152,
-    "producer_headroom": 7.502403014322032,
+    "gpu_update_seconds": 0.9503380681002553,
+    "producer_batches_per_second": 9.16692517730176,
+    "gpu_batches_per_second": 1.0522571215094223,
+    "producer_headroom": 8.711677963416546,
     "sustainable": true,
-    "buffer_seconds": 60.891601485301564,
+    "buffer_seconds": 60.82163635841634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 273708.2327722442,
-    "upload_bytes_per_second": 5415970.350410851,
+    "generation_bytes_per_second": 290061.3839081366,
+    "upload_bytes_per_second": 5380848.956825607,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.73141515136237,
+    "estimated_full_shards_per_hour": 62.240420703249676,
     "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": 1.3211888400001044,
-      "bytes": 9995304,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.8846473650010012,
       "bytes": 1315669,
@@ -562,6 +556,12 @@
       "bytes": 6766779,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.4542191169966827,
+      "bytes": 7891732,
+      "pages": 46,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -374,3 +374,4 @@
 {"step": 522400, "loss": 0.5688287019729614, "lr": 0.0003, "grad_norm": 0.17673751986089603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76461673859, "tokens_per_second": 146355.84917063563, "data_wait_s": 1.2958997103851289e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 522500, "loss": 0.6522858738899231, "lr": 0.0003, "grad_norm": 0.160575814897365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76476177656, "tokens_per_second": 151720.44219173345, "data_wait_s": 1.642899587750435e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 522600, "loss": 0.6110029816627502, "lr": 0.0003, "grad_norm": 0.17567874580709966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76490867880, "tokens_per_second": 153383.76898527102, "data_wait_s": 1.3652002962771803e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 522700, "loss": 0.5959990620613098, "lr": 0.0003, "grad_norm": 0.16184742550977668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76505539968, "tokens_per_second": 153816.60044779553, "data_wait_s": 1.134799822466448e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ffd00e0e9bf815a97b74dc50a9924f2b6ecf299630cd805deda7af7a476e9b3
-size 802014
+oid sha256:757942d33d6e5b276cefc1e422fdcedcc2a89e88e016c32dbf11b9c26bd7d083
+size 804132