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": 75385.93101545941,
-      "required_pages_per_second_upper_estimate": 0.0065342614365246545
+      "required_token_positions_per_second": 75402.8338973718,
+      "required_pages_per_second_upper_estimate": 0.006535726535488858
     },
     {
       "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": 9423.241376932427,
-      "required_pages_per_second_upper_estimate": 0.0037590718752722302
+      "required_token_positions_per_second": 9425.354237171476,
+      "required_pages_per_second_upper_estimate": 0.0037599147268116626
     },
     {
       "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": 18846.482753864853,
-      "required_pages_per_second_upper_estimate": 0.004359802098338675
+      "required_token_positions_per_second": 18850.70847434295,
+      "required_pages_per_second_upper_estimate": 0.004360779644401143
     },
     {
       "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": 28269.72413079728,
-      "required_pages_per_second_upper_estimate": 0.0017214778797857966
+      "required_token_positions_per_second": 28276.06271151443,
+      "required_pages_per_second_upper_estimate": 0.0017218638660955082
     },
     {
       "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": 28269.72413079728,
-      "required_pages_per_second_upper_estimate": 0.00217628161919375
+      "required_token_positions_per_second": 28276.06271151443,
+      "required_pages_per_second_upper_estimate": 0.002176769580683671
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28269.72413079728,
-      "required_pages_per_second_upper_estimate": 0.3510153610240918
+      "required_token_positions_per_second": 28276.06271151443,
+      "required_pages_per_second_upper_estimate": 0.35109406498273377
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08599026964034806,
-    "producer_p95_seconds": 0.2289254079987586,
-    "gpu_update_seconds": 0.9780074213699183,
-    "producer_batches_per_second": 11.629222750230607,
-    "gpu_batches_per_second": 1.022487128573397,
-    "producer_headroom": 11.373466154489426,
+    "producer_batch_seconds": 0.3041970867967905,
+    "producer_p95_seconds": 0.3055606399975659,
+    "gpu_update_seconds": 0.977788183668914,
+    "producer_batches_per_second": 3.2873424611985826,
+    "gpu_batches_per_second": 1.0227163885819743,
+    "producer_headroom": 3.214324614233059,
     "sustainable": true,
-    "buffer_seconds": 60.636460124934935,
+    "buffer_seconds": 60.62286738747267,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346965.0430858301,
+    "generation_bytes_per_second": 364130.4097933804,
     "upload_bytes_per_second": 5973834.3957768595,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.45062131339242,
+    "estimated_full_shards_per_hour": 78.1339094195467,
     "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": [

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -154,3 +154,4 @@
 {"step": 100400, "loss": 0.6271757539361715, "lr": 0.0003, "grad_norm": 0.1342398332251596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14692038012, "tokens_per_second": 150552.59828249892, "data_wait_s": 1.7460002709412947e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 100500, "loss": 0.609245615079999, "lr": 0.0003, "grad_norm": 0.1362154006841095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14706705037, "tokens_per_second": 148911.50150448031, "data_wait_s": 1.4929999451851472e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 100600, "loss": 0.6350348480045795, "lr": 0.0003, "grad_norm": 0.13767209344202713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14721221671, "tokens_per_second": 147239.6967303694, "data_wait_s": 1.650500053074211e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 100700, "loss": 0.6050354652106762, "lr": 0.0003, "grad_norm": 0.14852054598452386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14735771958, "tokens_per_second": 147169.68418247017, "data_wait_s": 1.1563999578356743e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d5556e2e64d4cd971ba1316e4f53a5084c203c8a8437ba64ba58f255f4ca07e
-size 332922
+oid sha256:9743012eed2644a53ffb8817d39a607fcb69c5b4f9ba96140f5eaa7dbc5fb2fa
+size 335040