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": 77272.31332951155,
-      "required_pages_per_second_upper_estimate": 0.005345877932604767
+      "required_token_positions_per_second": 77128.94176097361,
+      "required_pages_per_second_upper_estimate": 0.00533595915482544
     },
     {
       "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": 9659.039166188944,
-      "required_pages_per_second_upper_estimate": 0.004044037995209032
+      "required_token_positions_per_second": 9641.117720121701,
+      "required_pages_per_second_upper_estimate": 0.004036534659983027
     },
     {
       "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": 19318.07833237789,
-      "required_pages_per_second_upper_estimate": 0.004023446264649527
+      "required_token_positions_per_second": 19282.235440243403,
+      "required_pages_per_second_upper_estimate": 0.004015981135458544
     },
     {
       "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": 28977.11749856683,
-      "required_pages_per_second_upper_estimate": 0.0017371339485425584
+      "required_token_positions_per_second": 28923.353160365103,
+      "required_pages_per_second_upper_estimate": 0.001733910858560756
     },
     {
       "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": 28977.11749856683,
-      "required_pages_per_second_upper_estimate": 0.002252843428643752
+      "required_token_positions_per_second": 28923.353160365103,
+      "required_pages_per_second_upper_estimate": 0.002248663487832899
     },
     {
       "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": 28977.11749856683,
-      "required_pages_per_second_upper_estimate": 0.3716873500669159
+      "required_token_positions_per_second": 28923.353160365103,
+      "required_pages_per_second_upper_estimate": 0.37099771886411287
     }
   ],
   "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": 0.42092296799208384,
-    "producer_p95_seconds": 0.236458160001348,
-    "gpu_update_seconds": 0.9541321700257948,
-    "producer_batches_per_second": 2.3757316089693794,
-    "gpu_batches_per_second": 1.0480728261923766,
-    "producer_headroom": 2.266761955464827,
+    "producer_batch_seconds": 0.08498126613272916,
+    "producer_p95_seconds": 0.2372391749995586,
+    "gpu_update_seconds": 0.9559057639930638,
+    "producer_batches_per_second": 11.767299376760711,
+    "gpu_batches_per_second": 1.0461282248395942,
+    "producer_headroom": 11.248429300877552,
     "sustainable": true,
-    "buffer_seconds": 60.11032671162507,
+    "buffer_seconds": 60.22206313156302,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346994.7799779962,
-    "upload_bytes_per_second": 5206982.466853481,
+    "generation_bytes_per_second": 331244.8195139413,
+    "upload_bytes_per_second": 5243269.33049981,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.45700215821185,
+    "estimated_full_shards_per_hour": 71.07742728294066,
     "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": 1.1420412400002533,
-      "bytes": 10074433,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.157611688000543,
       "bytes": 7621104,
@@ -562,6 +556,12 @@
       "bytes": 10383198,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.2442956449995108,
+      "bytes": 13618701,
+      "pages": 45,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 624300, "loss": 0.5589475631713867, "lr": 0.0001, "grad_norm": 0.16092953620302844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91380728200, "tokens_per_second": 150561.9621450381, "data_wait_s": 1.2386000889819115e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 624400, "loss": 0.5014004111289978, "lr": 0.0001, "grad_norm": 0.15139213849634997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91395374412, "tokens_per_second": 152628.63471766797, "data_wait_s": 9.241000952897593e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 624500, "loss": 0.5853891968727112, "lr": 0.0001, "grad_norm": 0.18553739924719578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91410018711, "tokens_per_second": 151223.35202673485, "data_wait_s": 1.61259995365981e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 624600, "loss": 0.5893364548683167, "lr": 0.0001, "grad_norm": 0.15714717350095056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91424676083, "tokens_per_second": 151960.73589837409, "data_wait_s": 1.584900019224733e-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 sha256:5171b8ee0f7276c2b94ed2c1046042c0b96a19f986ec2db81f83ac9d577c658d
-size 429015
+oid sha256:608848aa6163c6668a529cb852283b541fd57685e4200be1147f55b389cb1dd5
+size 431133