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": 74383.26667979218,
-      "required_pages_per_second_upper_estimate": 0.006447353033138544
+      "required_token_positions_per_second": 75390.67366896351,
+      "required_pages_per_second_upper_estimate": 0.0065346725177102445
     },
     {
       "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": 9297.908334974023,
-      "required_pages_per_second_upper_estimate": 0.003709074650939055
+      "required_token_positions_per_second": 9423.83420862044,
+      "required_pages_per_second_upper_estimate": 0.003759308364696202
     },
     {
       "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": 18595.816669948046,
-      "required_pages_per_second_upper_estimate": 0.004301814911396925
+      "required_token_positions_per_second": 18847.66841724088,
+      "required_pages_per_second_upper_estimate": 0.004360076380693756
     },
     {
       "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": 27893.725004922067,
-      "required_pages_per_second_upper_estimate": 0.0016985815057349502
+      "required_token_positions_per_second": 28271.502625861314,
+      "required_pages_per_second_upper_estimate": 0.0017215861808041538
     },
     {
       "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": 27893.725004922067,
-      "required_pages_per_second_upper_estimate": 0.0021473361656516796
+      "required_token_positions_per_second": 28271.502625861314,
+      "required_pages_per_second_upper_estimate": 0.002176418532666969
     },
     {
       "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": 27893.725004922067,
-      "required_pages_per_second_upper_estimate": 0.3463467102688462
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28271.502625861314,
+      "required_pages_per_second_upper_estimate": 0.351037443980547
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.07125893121097704,
-    "producer_p95_seconds": 0.2031222629993863,
-    "gpu_update_seconds": 0.9911906708451916,
-    "producer_batches_per_second": 14.033328636929593,
-    "gpu_batches_per_second": 1.0088876231525632,
-    "producer_headroom": 13.90970442582928,
+    "producer_batch_seconds": 0.228715006718744,
+    "producer_p95_seconds": 0.196046645000024,
+    "gpu_update_seconds": 0.9779458971773588,
+    "producer_batches_per_second": 4.372253549718854,
+    "gpu_batches_per_second": 1.022551454928433,
+    "producer_headroom": 4.275827420366697,
     "sustainable": true,
-    "buffer_seconds": 60.46263092155669,
+    "buffer_seconds": 60.63264562499624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279919.0975107615,
-    "upload_bytes_per_second": 3715373.2403625427,
+    "generation_bytes_per_second": 279409.03349359956,
+    "upload_bytes_per_second": 3689120.873922546,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.064122142716734,
+    "estimated_full_shards_per_hour": 59.9546742783164,
     "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": 0.859392975999981,
-      "bytes": 2445063,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0558128229999966,
       "bytes": 6261983,
@@ -562,6 +556,12 @@
       "bytes": 10318034,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.292125334000957,
+      "bytes": 1081552,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -92,3 +92,4 @@
 {"step": 246300, "loss": 0.5327975749969482, "lr": 0.0003, "grad_norm": 0.15585395626792478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36051369840, "tokens_per_second": 147922.12861020927, "data_wait_s": 1.4101000488153659e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 246400, "loss": 0.6401682272553444, "lr": 0.0003, "grad_norm": 0.14270779245222198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36065781690, "tokens_per_second": 146995.02602726058, "data_wait_s": 1.0323999958927743e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 246500, "loss": 0.6035121232271194, "lr": 0.0003, "grad_norm": 0.13245181578642634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36080377744, "tokens_per_second": 148907.00192626932, "data_wait_s": 1.5286001143977046e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 246600, "loss": 0.5950597636401653, "lr": 0.0003, "grad_norm": 0.1482707455529183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36095092040, "tokens_per_second": 150548.3810541842, "data_wait_s": 1.485900065745227e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1dbe61cbb60cde278ee06f70812de1b77588e212c222312666104fdf9575079
-size 200823
+oid sha256:51e859312463287a7ff2880fccec1ac2b2db2a6b5603b2d7fead3c776a101246
+size 202941