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": 76765.82652480446,
-      "required_pages_per_second_upper_estimate": 0.00531083800024383
+      "required_token_positions_per_second": 76504.9110552513,
+      "required_pages_per_second_upper_estimate": 0.005292787262652845
     },
     {
       "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": 9595.728315600558,
-      "required_pages_per_second_upper_estimate": 0.004017531064148573
+      "required_token_positions_per_second": 9563.113881906413,
+      "required_pages_per_second_upper_estimate": 0.004003876081827657
     },
     {
       "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": 19191.456631201116,
-      "required_pages_per_second_upper_estimate": 0.003997074303533219
+      "required_token_positions_per_second": 19126.227763812825,
+      "required_pages_per_second_upper_estimate": 0.00398348885065715
     },
     {
       "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": 28787.184946801677,
-      "required_pages_per_second_upper_estimate": 0.001725747782074452
+      "required_token_positions_per_second": 28689.34164571924,
+      "required_pages_per_second_upper_estimate": 0.0017198822255726303
     },
     {
       "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": 28787.184946801677,
-      "required_pages_per_second_upper_estimate": 0.002238077008169016
+      "required_token_positions_per_second": 28689.34164571924,
+      "required_pages_per_second_upper_estimate": 0.0022304701218770558
     },
     {
       "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": 28787.184946801677,
-      "required_pages_per_second_upper_estimate": 0.36925109922655786
+      "required_token_positions_per_second": 28689.34164571924,
+      "required_pages_per_second_upper_estimate": 0.367996070416224
     }
   ],
   "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": 0.09009117356256979,
-    "producer_p95_seconds": 0.22987759600073332,
-    "gpu_update_seconds": 0.9604273586004719,
-    "producer_batches_per_second": 11.099866506961236,
-    "gpu_batches_per_second": 1.0412031592448523,
-    "producer_headroom": 10.660615470098627,
+    "producer_batch_seconds": 0.08495708125812484,
+    "producer_p95_seconds": 0.2440374170000723,
+    "gpu_update_seconds": 0.9637028392432763,
+    "producer_batches_per_second": 11.770649193581674,
+    "gpu_batches_per_second": 1.0376642666999147,
+    "producer_headroom": 11.34340804759124,
     "sustainable": true,
-    "buffer_seconds": 60.50692359182973,
+    "buffer_seconds": 60.713278872326406,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365043.024467226,
+    "generation_bytes_per_second": 366268.6460408284,
     "upload_bytes_per_second": 4816634.339576173,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.32973528397164,
+    "estimated_full_shards_per_hour": 78.59272514265669,
     "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": [

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

@@ -199,3 +199,4 @@
 {"step": 417000, "loss": 0.5681581497192383, "lr": 0.0003, "grad_norm": 0.15923104880186126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61037695917, "tokens_per_second": 150464.4131508963, "data_wait_s": 1.2131000403314829e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 417100, "loss": 0.5619052052497864, "lr": 0.0003, "grad_norm": 0.16201801202103625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61052323072, "tokens_per_second": 152051.67114879543, "data_wait_s": 7.458998879883438e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 417200, "loss": 0.5778757333755493, "lr": 0.0003, "grad_norm": 0.19271627092949623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61067065217, "tokens_per_second": 152987.1359427244, "data_wait_s": 1.6412999684689566e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 417300, "loss": 0.5162276029586792, "lr": 0.0003, "grad_norm": 0.15884595246507877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61081808801, "tokens_per_second": 153523.88054505913, "data_wait_s": 1.1416999768698588e-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 sha256:6549a40e6947a92d0a1d7fc355b9bfa4864d477d1d1f264c9df69c1a048cd790
-size 429015
+oid sha256:59beae7f163e6497f1c48c5fb53fe5c954a6fe5fa9c970d4599fad49773bf4de
+size 431133