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": 75149.15830674069,
-      "required_pages_per_second_upper_estimate": 0.0065137385782278595
+      "required_token_positions_per_second": 76080.61557549567,
+      "required_pages_per_second_upper_estimate": 0.0065944749335266764
     },
     {
       "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": 9393.644788342586,
-      "required_pages_per_second_upper_estimate": 0.0037472653535753097
+      "required_token_positions_per_second": 9510.076946936959,
+      "required_pages_per_second_upper_estimate": 0.0037937118824545074
     },
     {
       "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": 18787.289576685173,
-      "required_pages_per_second_upper_estimate": 0.004346108798562494
+      "required_token_positions_per_second": 19020.153893873918,
+      "required_pages_per_second_upper_estimate": 0.0043999778600721014
     },
     {
       "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": 28180.934365027755,
-      "required_pages_per_second_upper_estimate": 0.0017160710488943313
+      "required_token_positions_per_second": 28530.230840810877,
+      "required_pages_per_second_upper_estimate": 0.001737341371652547
     },
     {
       "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": 28180.934365027755,
-      "required_pages_per_second_upper_estimate": 0.0021694463372390095
+      "required_token_positions_per_second": 28530.230840810877,
+      "required_pages_per_second_upper_estimate": 0.0021963361468593243
     },
     {
       "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": 28180.934365027755,
-      "required_pages_per_second_upper_estimate": 0.34991288929346454
+      "required_token_positions_per_second": 28530.230840810877,
+      "required_pages_per_second_upper_estimate": 0.3542499825025873
     }
   ],
   "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": 0.27726213599197536,
-    "producer_p95_seconds": 0.381222374002391,
-    "gpu_update_seconds": 0.9810888326794046,
-    "producer_batches_per_second": 3.6066951458129948,
-    "gpu_batches_per_second": 1.019275693179534,
-    "producer_headroom": 3.538488330436146,
+    "producer_batch_seconds": 0.14641306001573184,
+    "producer_p95_seconds": 0.39043439400120405,
+    "gpu_update_seconds": 0.9690773325412813,
+    "producer_batches_per_second": 6.829991804641961,
+    "gpu_batches_per_second": 1.0319093909436805,
+    "producer_headroom": 6.6187902393212426,
     "sustainable": true,
-    "buffer_seconds": 60.82750762612309,
+    "buffer_seconds": 60.082794617559436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 453021.33522908384,
-    "upload_bytes_per_second": 5690868.20618875,
+    "generation_bytes_per_second": 461167.46831843315,
+    "upload_bytes_per_second": 5751632.416316283,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.2078327432097,
+    "estimated_full_shards_per_hour": 98.95580327191111,
     "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": 1.129864335001912,
-      "bytes": 7732628,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.990468955002143,
       "bytes": 4589915,
@@ -562,6 +556,12 @@
       "bytes": 997756,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 1.3090939850007999,
+      "bytes": 13205063,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -327,3 +327,4 @@
 {"step": 312900, "loss": 0.5546615142375231, "lr": 0.0003, "grad_norm": 0.1637472951903675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45799757670, "tokens_per_second": 150840.3689385695, "data_wait_s": 1.0062998626381159e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 313000, "loss": 0.6142807677388191, "lr": 0.0003, "grad_norm": 0.13964326787935719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45814553279, "tokens_per_second": 152406.05872717276, "data_wait_s": 9.756993677001446e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 313100, "loss": 0.6027722675353289, "lr": 0.0003, "grad_norm": 0.15013859134515675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45829217617, "tokens_per_second": 149519.503986863, "data_wait_s": 1.549800072098151e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 313200, "loss": 0.5805141348391771, "lr": 0.0003, "grad_norm": 0.1413036815673845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45843936515, "tokens_per_second": 151089.4814605454, "data_wait_s": 1.7166996258310974e-05, "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 sha256:7fc483a5a2dc2ba61e3b46a15e733f87514914f3b41aaf52e9f0ba2ae19d5dbe
-size 701685
+oid sha256:b2de09a5a3496775aa8f5af72b5647d63e3b9cb4b4000c3bc6d209940a01c122
+size 703803