Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76923.06776777981,
-      "required_pages_per_second_upper_estimate": 0.005321716314282815
+      "required_token_positions_per_second": 77046.24236086292,
+      "required_pages_per_second_upper_estimate": 0.005330237818436737
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9615.383470972476,
-      "required_pages_per_second_upper_estimate": 0.004025760267256478
+      "required_token_positions_per_second": 9630.780295107865,
+      "required_pages_per_second_upper_estimate": 0.004032206595999716
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19230.766941944952,
-      "required_pages_per_second_upper_estimate": 0.0040052616045785526
+      "required_token_positions_per_second": 19261.56059021573,
+      "required_pages_per_second_upper_estimate": 0.00401167510943024
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28846.15041291743,
-      "required_pages_per_second_upper_estimate": 0.0017292826717330376
+      "required_token_positions_per_second": 28892.340885323596,
+      "required_pages_per_second_upper_estimate": 0.00173205172002502
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28846.15041291743,
-      "required_pages_per_second_upper_estimate": 0.0022426613137978405
+      "required_token_positions_per_second": 28892.340885323596,
+      "required_pages_per_second_upper_estimate": 0.0022462524198570036
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28846.15041291743,
-      "required_pages_per_second_upper_estimate": 0.3700074449137059
+      "required_token_positions_per_second": 28892.340885323596,
+      "required_pages_per_second_upper_estimate": 0.370599926698267
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09345081030473068,
-    "producer_p95_seconds": 0.1861158070005331,
-    "gpu_update_seconds": 0.9584641140753086,
-    "producer_batches_per_second": 10.700816790556795,
-    "gpu_batches_per_second": 1.0433358800968398,
-    "producer_headroom": 10.256348885043206,
+    "producer_batch_seconds": 0.051604133328090995,
+    "producer_p95_seconds": 0.18105816900060745,
+    "gpu_update_seconds": 0.9569318079742136,
+    "producer_batches_per_second": 19.378292696869003,
+    "gpu_batches_per_second": 1.0450065424379194,
+    "producer_headroom": 18.543704665868354,
     "sustainable": true,
-    "buffer_seconds": 60.38323918674444,
+    "buffer_seconds": 60.28670390237546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350706.9521267033,
-    "upload_bytes_per_second": 4379228.935441446,
+    "generation_bytes_per_second": 350468.60353080445,
+    "upload_bytes_per_second": 4416879.550176273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.25354788637947,
+    "estimated_full_shards_per_hour": 75.20240382617092,
     "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": [
@@ -532,6 +532,12 @@
       "bytes": 1257250,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 0.9075711420000516,
+      "bytes": 7428003,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 447700, "loss": 0.641356348991394, "lr": 0.0003, "grad_norm": 0.1681395916240247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65529471295, "tokens_per_second": 150984.10910820056, "data_wait_s": 1.4199999895936344e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 447800, "loss": 0.5502520203590393, "lr": 0.0003, "grad_norm": 0.1577973275191342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65544005141, "tokens_per_second": 151587.22604370053, "data_wait_s": 7.556999662483577e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 447900, "loss": 0.5529193878173828, "lr": 0.0003, "grad_norm": 0.15696307109837762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65558740424, "tokens_per_second": 154257.8371211841, "data_wait_s": 1.1615000403253362e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 448000, "loss": 0.6013756990432739, "lr": 0.0003, "grad_norm": 0.16911388717877565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65573386473, "tokens_per_second": 152753.0935244055, "data_wait_s": 1.1083999197580852e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b50d4a00f3eabc05cc6f4df39a950daf8bd55a9abf157b1d266e7205913514a
-size 164034
+oid sha256:a0895ce2b3a0f5f104e978f090a58608333ef83ff4bcf4ad71009c38753016b6
+size 166152