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": 76924.88308471063,
-      "required_pages_per_second_upper_estimate": 0.005321841902120205
+      "required_token_positions_per_second": 76857.38466783531,
+      "required_pages_per_second_upper_estimate": 0.005317172205022866
     },
     {
       "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": 9615.610385588829,
-      "required_pages_per_second_upper_estimate": 0.004025855271667829
+      "required_token_positions_per_second": 9607.173083479414,
+      "required_pages_per_second_upper_estimate": 0.00402232274946552
     },
     {
       "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": 19231.220771177657,
-      "required_pages_per_second_upper_estimate": 0.004005356125239444
+      "required_token_positions_per_second": 19214.34616695883,
+      "required_pages_per_second_upper_estimate": 0.004001841590193901
     },
     {
       "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": 28846.831156766486,
-      "required_pages_per_second_upper_estimate": 0.0017293234812873531
+      "required_token_positions_per_second": 28821.519250438247,
+      "required_pages_per_second_upper_estimate": 0.0017278060711520182
     },
     {
       "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": 28846.831156766486,
-      "required_pages_per_second_upper_estimate": 0.002242714238637829
+      "required_token_positions_per_second": 28821.519250438247,
+      "required_pages_per_second_upper_estimate": 0.002240746349256118
     },
     {
       "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": 28846.831156766486,
-      "required_pages_per_second_upper_estimate": 0.37001617676487586
+      "required_token_positions_per_second": 28821.519250438247,
+      "required_pages_per_second_upper_estimate": 0.36969150280830476
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.5073274579607983,
-    "producer_p95_seconds": 0.34248151299834717,
-    "gpu_update_seconds": 0.9584414956966502,
-    "producer_batches_per_second": 1.9711134974233369,
-    "gpu_batches_per_second": 1.0433605019085102,
-    "producer_headroom": 1.8891969686582784,
+    "producer_batch_seconds": 0.11458010968755161,
+    "producer_p95_seconds": 0.3395944669973687,
+    "gpu_update_seconds": 0.9592832272219515,
+    "producer_batches_per_second": 8.727518264093995,
+    "gpu_batches_per_second": 1.042444996037263,
+    "producer_headroom": 8.372161886018612,
     "sustainable": true,
-    "buffer_seconds": 60.38181422888896,
+    "buffer_seconds": 60.43484331498294,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 270603.4737486303,
-    "upload_bytes_per_second": 5111100.11035178,
+    "generation_bytes_per_second": 283375.6438511428,
+    "upload_bytes_per_second": 5107965.243940179,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.06520613998587,
+    "estimated_full_shards_per_hour": 60.8058165230819,
     "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": 0.8855937030020868,
-      "bytes": 2335409,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0506481749980594,
-      "bytes": 6177700,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1661089699991862,
       "bytes": 11581137,
@@ -562,6 +550,18 @@
       "bytes": 5561171,
       "pages": 22,
       "shards": 2
+    },
+    {
+      "seconds": 1.2334552800020901,
+      "bytes": 12163400,
+      "pages": 48,
+      "shards": 3
+    },
+    {
+      "seconds": 1.5185914780013263,
+      "bytes": 255711,
+      "pages": 9,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -427,3 +427,4 @@
 {"step": 439800, "loss": 0.6283659338951111, "lr": 0.0003, "grad_norm": 0.1694328034722055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64373739148, "tokens_per_second": 149804.70826847616, "data_wait_s": 9.584997314959764e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 439900, "loss": 0.5995985269546509, "lr": 0.0003, "grad_norm": 0.1662764601544931, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64388436968, "tokens_per_second": 143802.86692522067, "data_wait_s": 1.1969001207035035e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 440000, "loss": 0.5777196884155273, "lr": 0.0003, "grad_norm": 0.15440248522207758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64403129477, "tokens_per_second": 152592.40903715484, "data_wait_s": 1.89220008905977e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 440100, "loss": 0.5686413645744324, "lr": 0.0003, "grad_norm": 0.15757820398411632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64417879158, "tokens_per_second": 140857.06358519776, "data_wait_s": 1.1529999028425664e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f5250a67860f7505e4251fda60f0b722df4608cdd7b077768238c3f0ee1be1d
-size 915834
+oid sha256:251ad52635d84a006036cad4ba266463550625e02849d9637477ee7a42f698c6
+size 917952