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": 75954.32085745948,
-      "required_pages_per_second_upper_estimate": 0.006583528027458336
+      "required_token_positions_per_second": 76172.36978296538,
+      "required_pages_per_second_upper_estimate": 0.006602427955681238
     },
     {
       "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": 9494.290107182434,
-      "required_pages_per_second_upper_estimate": 0.003787414276042139
+      "required_token_positions_per_second": 9521.546222870673,
+      "required_pages_per_second_upper_estimate": 0.0037982871481054225
     },
     {
       "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": 18988.58021436487,
-      "required_pages_per_second_upper_estimate": 0.004392673844995463
+      "required_token_positions_per_second": 19043.092445741346,
+      "required_pages_per_second_upper_estimate": 0.004405284290341906
     },
     {
       "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": 28482.870321547303,
-      "required_pages_per_second_upper_estimate": 0.0017344573645108362
+      "required_token_positions_per_second": 28564.638668612017,
+      "required_pages_per_second_upper_estimate": 0.0017394366278417155
     },
     {
       "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": 28482.870321547303,
-      "required_pages_per_second_upper_estimate": 0.0021926902029841063
+      "required_token_positions_per_second": 28564.638668612017,
+      "required_pages_per_second_upper_estimate": 0.0021989849566904198
     },
     {
       "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": 28482.870321547303,
-      "required_pages_per_second_upper_estimate": 0.35366192335879537
+      "required_token_positions_per_second": 28564.638668612017,
+      "required_pages_per_second_upper_estimate": 0.3546772125682856
     }
   ],
   "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": 0.36513871949216536,
-    "producer_p95_seconds": 0.2729201340007421,
-    "gpu_update_seconds": 0.9706886871961172,
-    "producer_batches_per_second": 2.7386851807740333,
-    "gpu_batches_per_second": 1.0301964091994829,
-    "producer_headroom": 2.6584107227690077,
+    "producer_batch_seconds": 0.06182848805471508,
+    "producer_p95_seconds": 0.21286628899906646,
+    "gpu_update_seconds": 0.9679100205241085,
+    "producer_batches_per_second": 16.17377412035453,
+    "gpu_batches_per_second": 1.033153887030238,
+    "producer_headroom": 15.654758040784648,
     "sustainable": true,
-    "buffer_seconds": 60.18269860615927,
+    "buffer_seconds": 60.010421272494725,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387973.0468708868,
-    "upload_bytes_per_second": 4002281.0740489964,
+    "generation_bytes_per_second": 388201.89586133236,
+    "upload_bytes_per_second": 3966952.9770159055,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.24998430819467,
+    "estimated_full_shards_per_hour": 83.29908997421244,
     "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": 1.1265525479993812,
-      "bytes": 6298017,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9624441889991431,
       "bytes": 1027701,
@@ -562,6 +556,12 @@
       "bytes": 13332712,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 0.7849141779988713,
+      "bytes": 1153413,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -117,3 +117,4 @@
 {"step": 291900, "loss": 0.6618529204279184, "lr": 0.0003, "grad_norm": 0.14905779807584116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42725726528, "tokens_per_second": 149057.4284896329, "data_wait_s": 1.3377000868786126e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 292000, "loss": 0.6027472764253616, "lr": 0.0003, "grad_norm": 0.1884136091430959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42740315641, "tokens_per_second": 149508.34723971388, "data_wait_s": 1.237399919773452e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 292100, "loss": 0.5817364305257797, "lr": 0.0003, "grad_norm": 0.15640337190866618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42755006915, "tokens_per_second": 151096.2911353824, "data_wait_s": 9.278999641537666e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 292200, "loss": 0.5701478384435177, "lr": 0.0003, "grad_norm": 0.15492571634711907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42769727323, "tokens_per_second": 150911.93648607002, "data_wait_s": 9.900999430101365e-06, "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:acf361e813d48976cfccfea660774430ca34cf84496609e2b6853ae99a3e3687
-size 253773
+oid sha256:a354bf00a8488385b7ee43b03d375ac0f6e18830080423cb590eeb854b60ba48
+size 255891