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": 77248.78630446953,
-      "required_pages_per_second_upper_estimate": 0.0053442502784221396
+      "required_token_positions_per_second": 76994.26246526305,
+      "required_pages_per_second_upper_estimate": 0.005326641728649163
     },
     {
       "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": 9656.09828805869,
-      "required_pages_per_second_upper_estimate": 0.004042806710948413
+      "required_token_positions_per_second": 9624.282808157881,
+      "required_pages_per_second_upper_estimate": 0.004029486233896714
     },
     {
       "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": 19312.19657611738,
-      "required_pages_per_second_upper_estimate": 0.004022221249932807
+      "required_token_positions_per_second": 19248.565616315762,
+      "required_pages_per_second_upper_estimate": 0.004008968599067384
     },
     {
       "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": 28968.29486417607,
-      "required_pages_per_second_upper_estimate": 0.001736605045082215
+      "required_token_positions_per_second": 28872.84842447364,
+      "required_pages_per_second_upper_estimate": 0.0017308831767603294
     },
     {
       "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": 28968.29486417607,
-      "required_pages_per_second_upper_estimate": 0.002252157507626538
+      "required_token_positions_per_second": 28872.84842447364,
+      "required_pages_per_second_upper_estimate": 0.0022447369667641937
     },
     {
       "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": 28968.29486417607,
-      "required_pages_per_second_upper_estimate": 0.371574182785958
+      "required_token_positions_per_second": 28872.84842447364,
+      "required_pages_per_second_upper_estimate": 0.3703498983398576
     }
   ],
   "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": 0.061402477453071924,
-    "producer_p95_seconds": 0.21938837799825706,
-    "gpu_update_seconds": 0.9544227621830504,
-    "producer_batches_per_second": 16.28598782132643,
-    "gpu_batches_per_second": 1.0477537204924794,
-    "producer_headroom": 15.54371748130989,
+    "producer_batch_seconds": 0.09693855689835118,
+    "producer_p95_seconds": 0.2297605530002329,
+    "gpu_update_seconds": 0.9575778459240822,
+    "producer_batches_per_second": 10.315812737429031,
+    "gpu_batches_per_second": 1.0443015199824088,
+    "producer_headroom": 9.878193740063502,
     "sustainable": true,
-    "buffer_seconds": 60.12863401753218,
+    "buffer_seconds": 60.32740429321718,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389321.88874675933,
-    "upload_bytes_per_second": 3598730.5110698557,
+    "generation_bytes_per_second": 404890.64349356346,
+    "upload_bytes_per_second": 3538129.8807488834,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.53941437532507,
+    "estimated_full_shards_per_hour": 86.88010672192743,
     "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": 1.532587591000265,
-      "bytes": 13240452,
-      "pages": 31,
-      "shards": 4
-    },
     {
       "seconds": 1.8508207049999328,
       "bytes": 10873151,
@@ -562,6 +556,12 @@
       "bytes": 5864353,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.5614973880001344,
+      "bytes": 5966551,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -181,3 +181,4 @@
 {"step": 943100, "loss": 0.5516357421875, "lr": 3e-05, "grad_norm": 0.3265654176664102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138043429903, "tokens_per_second": 152675.90930268308, "data_wait_s": 1.3322001905180514e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 943200, "loss": 0.5323226451873779, "lr": 3e-05, "grad_norm": 0.34239969708404827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138057994585, "tokens_per_second": 151893.92452867344, "data_wait_s": 1.3779997971141711e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 943300, "loss": 0.5704653859138489, "lr": 3e-05, "grad_norm": 0.3414814722542913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138072732208, "tokens_per_second": 153684.25294178058, "data_wait_s": 1.7148002370959148e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 943400, "loss": 0.4970327913761139, "lr": 3e-05, "grad_norm": 0.341055058838513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138087498363, "tokens_per_second": 143743.3284748618, "data_wait_s": 1.3756001862930134e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b2b9ff0560d7d814daf9a2933c9928a7f3cf92699daa4f3ade96d8d216090e90
-size 390108
+oid sha256:33e83b61b1b4cd9133ba42b5d5ce46a4af56964b7bac8eda8fa8124539ee22fc
+size 392226