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": 73641.73286784852,
-      "required_pages_per_second_upper_estimate": 0.0063830787617195316
+      "required_token_positions_per_second": 75269.67977556879,
+      "required_pages_per_second_upper_estimate": 0.006524185073687009
     },
     {
       "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": 9205.216608481065,
-      "required_pages_per_second_upper_estimate": 0.003672098535376203
+      "required_token_positions_per_second": 9408.709971946098,
+      "required_pages_per_second_upper_estimate": 0.003753275080559318
     },
     {
       "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": 18410.43321696213,
-      "required_pages_per_second_upper_estimate": 0.004258929712185973
+      "required_token_positions_per_second": 18817.419943892197,
+      "required_pages_per_second_upper_estimate": 0.004353078928739488
     },
     {
       "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": 27615.649825443197,
-      "required_pages_per_second_upper_estimate": 0.0016816481862524002
+      "required_token_positions_per_second": 28226.129915838294,
+      "required_pages_per_second_upper_estimate": 0.0017188232208159638
     },
     {
       "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": 27615.649825443197,
-      "required_pages_per_second_upper_estimate": 0.0021259291685740327
+      "required_token_positions_per_second": 28226.129915838294,
+      "required_pages_per_second_upper_estimate": 0.002172925615849697
     },
     {
       "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": 27615.649825443197,
-      "required_pages_per_second_upper_estimate": 0.3428939471974769
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28226.129915838294,
+      "required_pages_per_second_upper_estimate": 0.3504740667747531
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.17912464750799018,
-    "producer_p95_seconds": 0.30468093899980886,
-    "gpu_update_seconds": 1.0011714435387646,
-    "producer_batches_per_second": 5.582704635638673,
-    "gpu_batches_per_second": 0.9988299271355323,
-    "producer_headroom": 5.5892444589129235,
+    "producer_batch_seconds": 0.06804252624212381,
+    "producer_p95_seconds": 0.22329473100035102,
+    "gpu_update_seconds": 0.9795179177038403,
+    "producer_batches_per_second": 14.696691249257576,
+    "gpu_batches_per_second": 1.0209103702198457,
+    "producer_headroom": 14.395672409609032,
     "sustainable": true,
-    "buffer_seconds": 60.07028661232588,
+    "buffer_seconds": 60.7301108976381,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 496838.0977403242,
-    "upload_bytes_per_second": 5753543.420760775,
+    "generation_bytes_per_second": 499545.73632224073,
+    "upload_bytes_per_second": 5796321.512549307,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.6098899760942,
+    "estimated_full_shards_per_hour": 107.19088618517318,
     "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": 1.0756502530002763,
-      "bytes": 6680691,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0957601480004087,
       "bytes": 10587203,
@@ -562,6 +556,12 @@
       "bytes": 7611195,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.260750511999504,
+      "bytes": 10895855,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -160,3 +160,4 @@
 {"step": 371000, "loss": 0.5991155337542295, "lr": 0.0003, "grad_norm": 0.1632551389304045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54303601846, "tokens_per_second": 148165.97946345946, "data_wait_s": 1.1613003152888268e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 371100, "loss": 0.5825608931481838, "lr": 0.0003, "grad_norm": 0.13730084919229854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54318136898, "tokens_per_second": 147243.17809376516, "data_wait_s": 1.3117001799400896e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 371200, "loss": 0.6160933915525675, "lr": 0.0003, "grad_norm": 0.16668968304273502, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54332732499, "tokens_per_second": 148870.62071015942, "data_wait_s": 1.2242999218869954e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 371300, "loss": 0.66576480679214, "lr": 0.0003, "grad_norm": 0.19301529987524496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54347462097, "tokens_per_second": 149414.58814127877, "data_wait_s": 1.560300006531179e-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 sha256:5eb1bac3276c5ce2f3107ff5e0769921db08073925fa87a3217d6ffe5db33ed7
-size 345630
+oid sha256:8940263d90a34dab5ca58985dd11f9cada7b8262269abe60e554263a601d50c1
+size 347748