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": 76092.6606814469,
-      "required_pages_per_second_upper_estimate": 0.006595518972782487
+      "required_token_positions_per_second": 75952.4825342208,
+      "required_pages_per_second_upper_estimate": 0.006583368686259195
     },
     {
       "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": 9511.582585180862,
-      "required_pages_per_second_upper_estimate": 0.003794312504061298
+      "required_token_positions_per_second": 9494.0603167776,
+      "required_pages_per_second_upper_estimate": 0.0037873226092139784
     },
     {
       "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": 19023.165170361724,
-      "required_pages_per_second_upper_estimate": 0.004400674465890899
+      "required_token_positions_per_second": 18988.1206335552,
+      "required_pages_per_second_upper_estimate": 0.004392567529063493
     },
     {
       "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": 28534.747755542587,
-      "required_pages_per_second_upper_estimate": 0.001737616428061288
+      "required_token_positions_per_second": 28482.180950332804,
+      "required_pages_per_second_upper_estimate": 0.0017344153854207246
     },
     {
       "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": 28534.747755542587,
-      "required_pages_per_second_upper_estimate": 0.002196683871458992
+      "required_token_positions_per_second": 28482.180950332804,
+      "required_pages_per_second_upper_estimate": 0.0021926371332797127
     },
     {
       "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": 28534.747755542587,
-      "required_pages_per_second_upper_estimate": 0.354306067466414
+      "required_token_positions_per_second": 28482.180950332804,
+      "required_pages_per_second_upper_estimate": 0.35365336367548833
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.1796711110623903,
+    "producer_batch_seconds": 0.08103640543748725,
     "producer_p95_seconds": 0.29974260100061656,
-    "gpu_update_seconds": 0.9689239322127757,
-    "producer_batches_per_second": 5.565725029956278,
-    "gpu_batches_per_second": 1.032072763148965,
-    "producer_headroom": 5.392764181640305,
+    "gpu_update_seconds": 0.9707121813534066,
+    "producer_batches_per_second": 12.340132741591255,
+    "gpu_batches_per_second": 1.0301714753447917,
+    "producer_headroom": 11.97871717178064,
     "sustainable": true,
-    "buffer_seconds": 60.0732837971921,
+    "buffer_seconds": 60.18415524391121,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 155,
+    "upload_pages_per_commit": 159,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 561797.2639199487,
+    "generation_bytes_per_second": 563978.0897682094,
     "upload_bytes_per_second": 4954328.8602289,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.27430743312286,
+    "estimated_full_shards_per_hour": 60.508284663127476,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 381400, "loss": 0.5771404895931482, "lr": 0.0003, "grad_norm": 0.14902781861344497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55825602854, "tokens_per_second": 148512.8178592142, "data_wait_s": 9.18500154512003e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 381500, "loss": 0.5936720445752144, "lr": 0.0003, "grad_norm": 0.15501056398168583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55840213513, "tokens_per_second": 149718.14063646214, "data_wait_s": 1.219599653268233e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 381600, "loss": 0.573487963527441, "lr": 0.0003, "grad_norm": 0.1656134459225898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55854726840, "tokens_per_second": 147647.3880881195, "data_wait_s": 1.3763001334154978e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 381700, "loss": 0.613114608451724, "lr": 0.0003, "grad_norm": 0.15971321352573042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55869369087, "tokens_per_second": 149673.83336199875, "data_wait_s": 1.0146999557036906e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:942fe422dd756c5361c73133c5941f6366c07c2596d219beaedda43535fd451e
-size 567468
+oid sha256:d98c12469fa1d707d2aebdee0c1c1cc8c616bcb9698cf6ae1abac8bd2533ed79
+size 569586