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": 75082.21784478199,
-      "required_pages_per_second_upper_estimate": 0.006507936348644329
+      "required_token_positions_per_second": 75712.41883755567,
+      "required_pages_per_second_upper_estimate": 0.006562560573468138
     },
     {
       "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": 9385.277230597749,
-      "required_pages_per_second_upper_estimate": 0.003743927409684757
+      "required_token_positions_per_second": 9464.052354694459,
+      "required_pages_per_second_upper_estimate": 0.003775351984479998
     },
     {
       "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": 18770.554461195497,
-      "required_pages_per_second_upper_estimate": 0.004342237424121931
+      "required_token_positions_per_second": 18928.104709388917,
+      "required_pages_per_second_upper_estimate": 0.004378683901251813
     },
     {
       "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": 28155.831691793246,
-      "required_pages_per_second_upper_estimate": 0.0017145424277978975
+      "required_token_positions_per_second": 28392.157064083374,
+      "required_pages_per_second_upper_estimate": 0.0017289334030669626
     },
     {
       "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": 28155.831691793246,
-      "required_pages_per_second_upper_estimate": 0.00216751386396477
+      "required_token_positions_per_second": 28392.157064083374,
+      "required_pages_per_second_upper_estimate": 0.0021857068453141645
     },
     {
       "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": 28155.831691793246,
-      "required_pages_per_second_upper_estimate": 0.34960119810513485
+      "required_token_positions_per_second": 28392.157064083374,
+      "required_pages_per_second_upper_estimate": 0.35253556829883625
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.1856948795781932,
-    "producer_p95_seconds": 0.2789090709993616,
-    "gpu_update_seconds": 0.9819635343273746,
-    "producer_batches_per_second": 5.38517810653425,
-    "gpu_batches_per_second": 1.0183677550561794,
-    "producer_headroom": 5.288048526474771,
+    "producer_batch_seconds": 0.08917709082825809,
+    "producer_p95_seconds": 0.2459568300000683,
+    "gpu_update_seconds": 0.973790048343148,
+    "producer_batches_per_second": 11.213642323518407,
+    "gpu_batches_per_second": 1.0269154030701453,
+    "producer_headroom": 10.91973330032176,
     "sustainable": true,
-    "buffer_seconds": 60.88173912829723,
+    "buffer_seconds": 60.374982997275175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 151,
+    "upload_pages_per_commit": 158,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 621632.3561263169,
-    "upload_bytes_per_second": 5739472.249607404,
+    "generation_bytes_per_second": 607219.8030335731,
+    "upload_bytes_per_second": 5808645.32920419,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.69391638203683,
+    "estimated_full_shards_per_hour": 65.14761718871782,
     "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": 1.0041788279995671,
-      "bytes": 4770921,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.17236374300046,
       "bytes": 5601597,
@@ -562,6 +556,12 @@
       "bytes": 13335425,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.1053190409984381,
+      "bytes": 10451885,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -166,3 +166,4 @@
 {"step": 371600, "loss": 0.5862086229026318, "lr": 0.0003, "grad_norm": 0.18389443548017723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54391519551, "tokens_per_second": 150760.67801553255, "data_wait_s": 1.5630001144018024e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 371700, "loss": 0.6007288508117199, "lr": 0.0003, "grad_norm": 0.1356210092778565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54406064263, "tokens_per_second": 147811.33434016505, "data_wait_s": 8.936000085668638e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 371800, "loss": 0.5621264316141605, "lr": 0.0003, "grad_norm": 0.14949414118034993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54420547782, "tokens_per_second": 148312.06734920602, "data_wait_s": 8.563998562749475e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 371900, "loss": 0.5916075948625803, "lr": 0.0003, "grad_norm": 0.1549158219388106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54435179152, "tokens_per_second": 148884.56980442794, "data_wait_s": 1.1111998901469633e-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:ed3fdae93c057cdf84d8cf17e4fa1c6c9b8ec3b957d5fe7ca7f7795b7f8c96e5
-size 358338
+oid sha256:0df5801f0a0daa8ac56710b207a929fa087ca85774145981320c139f3a55e62a
+size 360456