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": 76787.61513387342,
-      "required_pages_per_second_upper_estimate": 0.005312345386775759
+      "required_token_positions_per_second": 76809.76409613734,
+      "required_pages_per_second_upper_estimate": 0.005313877703377851
     },
     {
       "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": 9598.451891734177,
-      "required_pages_per_second_upper_estimate": 0.004018671368600983
+      "required_token_positions_per_second": 9601.220512017167,
+      "required_pages_per_second_upper_estimate": 0.004019830532098105
     },
     {
       "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": 19196.903783468355,
-      "required_pages_per_second_upper_estimate": 0.00399820880169942
+      "required_token_positions_per_second": 19202.441024034335,
+      "required_pages_per_second_upper_estimate": 0.00399936206288246
     },
     {
       "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": 28795.355675202532,
-      "required_pages_per_second_upper_estimate": 0.0017262376047661033
+      "required_token_positions_per_second": 28803.661536051502,
+      "required_pages_per_second_upper_estimate": 0.001726735528441684
     },
     {
       "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": 28795.355675202532,
-      "required_pages_per_second_upper_estimate": 0.002238712246362947
+      "required_token_positions_per_second": 28803.661536051502,
+      "required_pages_per_second_upper_estimate": 0.002239357990510333
     },
     {
       "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": 28795.355675202532,
-      "required_pages_per_second_upper_estimate": 0.3693559045574394
+      "required_token_positions_per_second": 28803.661536051502,
+      "required_pages_per_second_upper_estimate": 0.3694624432222714
     }
   ],
   "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": 0.0602029283751051,
-    "producer_p95_seconds": 0.19551041100021394,
-    "gpu_update_seconds": 0.9601548357956005,
-    "producer_batches_per_second": 16.610487678760762,
-    "gpu_batches_per_second": 1.0414986861690732,
-    "producer_headroom": 15.948640069685384,
+    "producer_batch_seconds": 0.07212252820315257,
+    "producer_p95_seconds": 0.20351836500049103,
+    "gpu_update_seconds": 0.9598779643134939,
+    "producer_batches_per_second": 13.865293201913694,
+    "gpu_batches_per_second": 1.0417991006963072,
+    "producer_headroom": 13.308989413262642,
     "sustainable": true,
-    "buffer_seconds": 60.48975465512283,
+    "buffer_seconds": 60.47231175175012,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295988.0364706684,
-    "upload_bytes_per_second": 5500821.607422729,
+    "generation_bytes_per_second": 295503.61445087806,
+    "upload_bytes_per_second": 5453448.462381708,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.512142377758394,
+    "estimated_full_shards_per_hour": 63.40819668907887,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 12102575,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 0.872640274999867,
+      "bytes": 1185922,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -111,3 +111,4 @@
 {"step": 771100, "loss": 0.4970531463623047, "lr": 3e-05, "grad_norm": 0.23643469945190743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112866569411, "tokens_per_second": 150266.1260719092, "data_wait_s": 9.981000403058715e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 771200, "loss": 0.5060374736785889, "lr": 3e-05, "grad_norm": 0.25702834694377064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112881244233, "tokens_per_second": 152751.98276192133, "data_wait_s": 1.2203001460875385e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 771300, "loss": 0.5389420986175537, "lr": 3e-05, "grad_norm": 0.24801907482809776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112895902196, "tokens_per_second": 152771.32929801662, "data_wait_s": 8.81000050867442e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 771400, "loss": 0.5066722631454468, "lr": 3e-05, "grad_norm": 0.24753382743820754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112910503453, "tokens_per_second": 152183.82168502882, "data_wait_s": 1.6093999875010923e-05, "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 sha256:3503f744a4dc117428416b0716f29585697e8294d4942558e89a9a8a63202a02
-size 241065
+oid sha256:f0f4a4042610923bbe0622fdce747c09c3e0ec9e68094e3f84c576108cfbc3d7
+size 243183