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": 75809.44552450252,
-      "required_pages_per_second_upper_estimate": 0.006570970600780801
+      "required_token_positions_per_second": 75813.08876281223,
+      "required_pages_per_second_upper_estimate": 0.006571286387443784
     },
     {
       "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": 9476.180690562815,
-      "required_pages_per_second_upper_estimate": 0.003780190158992666
+      "required_token_positions_per_second": 9476.636095351529,
+      "required_pages_per_second_upper_estimate": 0.0037803718267717923
     },
     {
       "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": 18952.36138112563,
-      "required_pages_per_second_upper_estimate": 0.004384295255355259
+      "required_token_positions_per_second": 18953.272190703057,
+      "required_pages_per_second_upper_estimate": 0.004384505955121296
     },
     {
       "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": 28428.542071688444,
-      "required_pages_per_second_upper_estimate": 0.00173114905913273
+      "required_token_positions_per_second": 28429.908286054586,
+      "required_pages_per_second_upper_estimate": 0.0017312322544196551
     },
     {
       "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": 28428.542071688444,
-      "required_pages_per_second_upper_estimate": 0.0021885078639197505
+      "required_token_positions_per_second": 28429.908286054586,
+      "required_pages_per_second_upper_estimate": 0.0021886130388835794
     },
     {
       "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": 28428.542071688444,
-      "required_pages_per_second_upper_estimate": 0.35298734832050416
+      "required_token_positions_per_second": 28429.908286054586,
+      "required_pages_per_second_upper_estimate": 0.3530043121305063
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11607938136737062,
-    "producer_p95_seconds": 0.30987315899983514,
-    "gpu_update_seconds": 0.9725437178691702,
-    "producer_batches_per_second": 8.614794360724389,
-    "gpu_batches_per_second": 1.0282314117364164,
-    "producer_headroom": 8.378264136257258,
+    "producer_batch_seconds": 0.08732361685946444,
+    "producer_p95_seconds": 0.30789627699778066,
+    "gpu_update_seconds": 0.9724969817634577,
+    "producer_batches_per_second": 11.45165576008338,
+    "gpu_batches_per_second": 1.0282808263185252,
+    "producer_headroom": 11.136700662875201,
     "sustainable": true,
-    "buffer_seconds": 60.29771050788855,
+    "buffer_seconds": 60.29481286933438,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287816.8991405864,
-    "upload_bytes_per_second": 3615993.4112310526,
+    "generation_bytes_per_second": 311545.0810315794,
+    "upload_bytes_per_second": 3403538.95858955,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.7588065210647,
+    "estimated_full_shards_per_hour": 66.85032199106728,
     "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": 1.263599650001197,
-      "bytes": 13782793,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 1.1351800550000917,
       "bytes": 9521447,
@@ -562,6 +556,12 @@
       "bytes": 5800923,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 6.862793137002882,
+      "bytes": 7606529,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 295100, "loss": 0.6096308417618275, "lr": 0.0003, "grad_norm": 0.17576634185582596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43194600819, "tokens_per_second": 133534.60692269387, "data_wait_s": 1.0063999070553109e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 295200, "loss": 0.6159439757466316, "lr": 0.0003, "grad_norm": 0.14220817293968846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43209262245, "tokens_per_second": 149883.7040387938, "data_wait_s": 1.4658002328360453e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 295300, "loss": 0.6048957770690322, "lr": 0.0003, "grad_norm": 0.16730942682486583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43223904710, "tokens_per_second": 149659.93292838745, "data_wait_s": 1.556899951538071e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 295400, "loss": 0.586178058758378, "lr": 0.0003, "grad_norm": 0.1444518754760649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43238613010, "tokens_per_second": 150779.14276032688, "data_wait_s": 8.979000995168462e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:018d778fad43fb74ef3d234ebf2fd73f5c53a4b74107c7e61d7087711567571d
-size 322332
+oid sha256:603a3daa655871ffdb38cd7b6eaa1e446df5eda153d5497822fa4859c3772f8a
+size 324450