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": 76961.9928842833,
-      "required_pages_per_second_upper_estimate": 0.00532440924416124
+      "required_token_positions_per_second": 77045.2394296334,
+      "required_pages_per_second_upper_estimate": 0.005330168433327158
     },
     {
       "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": 9620.249110535413,
-      "required_pages_per_second_upper_estimate": 0.004027797408935371
+      "required_token_positions_per_second": 9630.654928704174,
+      "required_pages_per_second_upper_estimate": 0.004032154107704439
     },
     {
       "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": 19240.498221070826,
-      "required_pages_per_second_upper_estimate": 0.004007288373389384
+      "required_token_positions_per_second": 19261.30985740835,
+      "required_pages_per_second_upper_estimate": 0.004011622888398732
     },
     {
       "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": 28860.74733160624,
-      "required_pages_per_second_upper_estimate": 0.0017301577347202275
+      "required_token_positions_per_second": 28891.964786112523,
+      "required_pages_per_second_upper_estimate": 0.0017320291734515864
     },
     {
       "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": 28860.74733160624,
-      "required_pages_per_second_upper_estimate": 0.0022437961600208357
+      "required_token_positions_per_second": 28891.964786112523,
+      "required_pages_per_second_upper_estimate": 0.0022462231797976285
     },
     {
       "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": 28860.74733160624,
-      "required_pages_per_second_upper_estimate": 0.3701946785136958
+      "required_token_positions_per_second": 28891.964786112523,
+      "required_pages_per_second_upper_estimate": 0.37059510250141126
     }
   ],
   "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": 0.0642742727890635,
-    "producer_p95_seconds": 0.16281288800018956,
-    "gpu_update_seconds": 0.9579793510655864,
-    "producer_batches_per_second": 15.5583246080717,
-    "gpu_batches_per_second": 1.043863835778582,
-    "producer_headroom": 14.904553711708273,
+    "producer_batch_seconds": 0.09461671117188075,
+    "producer_p95_seconds": 0.1649834190002366,
+    "gpu_update_seconds": 0.9569442647697516,
+    "producer_batches_per_second": 10.568957508821033,
+    "gpu_batches_per_second": 1.0449929393125188,
+    "producer_headroom": 10.113903272661489,
     "sustainable": true,
-    "buffer_seconds": 60.35269911713194,
+    "buffer_seconds": 60.28748868049435,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 499399.90262217267,
-    "upload_bytes_per_second": 4940242.496369626,
+    "generation_bytes_per_second": 526300.8683167761,
+    "upload_bytes_per_second": 4716074.82304089,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.15959366797338,
+    "estimated_full_shards_per_hour": 112.93191468360388,
     "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": [
@@ -226,6 +226,12 @@
       "bytes": 6005067,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9762548340004287,
+      "bytes": 2290276,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 764400, "loss": 0.5274600982666016, "lr": 3e-05, "grad_norm": 0.3359093690390421, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111885697494, "tokens_per_second": 151168.8128846808, "data_wait_s": 9.989999853132758e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 764500, "loss": 0.5336703658103943, "lr": 3e-05, "grad_norm": 0.25384554859239883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111900240353, "tokens_per_second": 150794.78291010932, "data_wait_s": 1.669200082687894e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 764600, "loss": 0.4760446846485138, "lr": 3e-05, "grad_norm": 0.2547620255634237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111914677661, "tokens_per_second": 150327.97626785203, "data_wait_s": 8.896000508684665e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 764700, "loss": 0.5431162714958191, "lr": 3e-05, "grad_norm": 0.2525561752895532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111929382897, "tokens_per_second": 152351.71289198974, "data_wait_s": 8.695999895280693e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7eea1324808a46e297c41ac68944ca59677cd5f675f5006832cda7b2c417af2
-size 97593
+oid sha256:92c93f0123921b37d7b1d91d24957c38ccf372048499e21d782333268848386f
+size 99711