Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74928.87996466512,
-      "required_pages_per_second_upper_estimate": 0.006494645409827102
+      "required_token_positions_per_second": 75027.15291354906,
+      "required_pages_per_second_upper_estimate": 0.00650316345996585
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9366.10999558314,
-      "required_pages_per_second_upper_estimate": 0.0037362813130617277
+      "required_token_positions_per_second": 9378.394114193632,
+      "required_pages_per_second_upper_estimate": 0.0037411816316393937
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18732.21999116628,
-      "required_pages_per_second_upper_estimate": 0.004333369419144301
+      "required_token_positions_per_second": 18756.788228387264,
+      "required_pages_per_second_upper_estimate": 0.004339052848439169
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28098.329986749417,
-      "required_pages_per_second_upper_estimate": 0.001711040875648859
+      "required_token_positions_per_second": 28135.182342580894,
+      "required_pages_per_second_upper_estimate": 0.0017132849907696278
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28098.329986749417,
-      "required_pages_per_second_upper_estimate": 0.002163087223535589
+      "required_token_positions_per_second": 28135.182342580894,
+      "required_pages_per_second_upper_estimate": 0.0021659242199013463
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28098.329986749417,
-      "required_pages_per_second_upper_estimate": 0.3488872193743176
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28135.182342580894,
+      "required_pages_per_second_upper_estimate": 0.3493448022968436
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.23290687447651237,
-    "producer_p95_seconds": 0.23780862400053593,
-    "gpu_update_seconds": 0.9839730693261207,
-    "producer_batches_per_second": 4.2935615457793,
-    "gpu_batches_per_second": 1.0162879769512954,
-    "producer_headroom": 4.224748932541062,
+    "producer_batch_seconds": 0.08311125832030086,
+    "producer_p95_seconds": 0.21558112999991863,
+    "gpu_update_seconds": 0.982684230134042,
+    "producer_batches_per_second": 12.032064249901252,
+    "gpu_batches_per_second": 1.0176208891269132,
+    "producer_headroom": 11.823719794337542,
     "sustainable": true,
-    "buffer_seconds": 60.02235722889336,
+    "buffer_seconds": 60.92642226831061,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353995.5198559459,
-    "upload_bytes_per_second": 5275946.098758672,
+    "generation_bytes_per_second": 353023.5112007192,
+    "upload_bytes_per_second": 5385865.091543153,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.95919796713622,
+    "estimated_full_shards_per_hour": 75.75062753692801,
     "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": 1.3317575169999145,
-      "bytes": 24816,
-      "pages": 1,
-      "shards": 1
-    },
     {
       "seconds": 1.540608198999962,
       "bytes": 9815303,
@@ -562,6 +556,12 @@
       "bytes": 9572867,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.2363076699984958,
+      "bytes": 8177813,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -106,3 +106,4 @@
 {"step": 205600, "loss": 0.5864623375236988, "lr": 0.0003, "grad_norm": 0.17855335037674847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30092330583, "tokens_per_second": 148678.9822647812, "data_wait_s": 1.805900137696881e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 205700, "loss": 0.6235309913754463, "lr": 0.0003, "grad_norm": 0.149813195949142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30107028663, "tokens_per_second": 147879.35779941638, "data_wait_s": 1.8307000573258847e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 205800, "loss": 0.5915245972573757, "lr": 0.0003, "grad_norm": 0.14105706562464318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30121589290, "tokens_per_second": 146875.43034318462, "data_wait_s": 1.649799924052786e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 205900, "loss": 0.5897836480289698, "lr": 0.0003, "grad_norm": 0.17344364810716098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30136072765, "tokens_per_second": 145956.09273378475, "data_wait_s": 1.3511998986359686e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d22591bce72f59b7c3417f65f8e327ddedb536903db8bea4f2f701b852bf7b44
-size 230475
+oid sha256:f10ea02a0bd8bd73d63450400308eb323d62097c9f68bf4a742260f5ada73a62
+size 232593