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": 76801.26090079229,
-      "required_pages_per_second_upper_estimate": 0.005313289432593748
+      "required_token_positions_per_second": 76955.28194169965,
+      "required_pages_per_second_upper_estimate": 0.005323944965581771
     },
     {
       "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": 9600.157612599036,
-      "required_pages_per_second_upper_estimate": 0.004019385518307597
+      "required_token_positions_per_second": 9619.410242712456,
+      "required_pages_per_second_upper_estimate": 0.0040274461924954515
     },
     {
       "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": 19200.315225198072,
-      "required_pages_per_second_upper_estimate": 0.003998919315045952
+      "required_token_positions_per_second": 19238.820485424912,
+      "required_pages_per_second_upper_estimate": 0.004006938945299204
     },
     {
       "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": 28800.472837797108,
-      "required_pages_per_second_upper_estimate": 0.0017265443708501932
+      "required_token_positions_per_second": 28858.23072813737,
+      "required_pages_per_second_upper_estimate": 0.0017300068681849
     },
     {
       "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": 28800.472837797108,
-      "required_pages_per_second_upper_estimate": 0.0022391100832466555
+      "required_token_positions_per_second": 28858.23072813737,
+      "required_pages_per_second_upper_estimate": 0.0022436005051704997
     },
     {
       "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": 28800.472837797108,
-      "required_pages_per_second_upper_estimate": 0.36942154202482147
+      "required_token_positions_per_second": 28858.23072813737,
+      "required_pages_per_second_upper_estimate": 0.3701623982265154
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11158727382036204,
-    "producer_p95_seconds": 0.18810637299975497,
-    "gpu_update_seconds": 0.9599842389988602,
-    "producer_batches_per_second": 8.961595402087184,
-    "gpu_batches_per_second": 1.041683768728194,
-    "producer_headroom": 8.602990342288349,
+    "producer_batch_seconds": 0.16534031250003522,
+    "producer_p95_seconds": 0.18688368299990543,
+    "gpu_update_seconds": 0.9580628923672244,
+    "producer_batches_per_second": 6.048131788790389,
+    "gpu_batches_per_second": 1.0437728127943204,
+    "producer_headroom": 5.794490634986675,
     "sustainable": true,
-    "buffer_seconds": 60.4790070569282,
+    "buffer_seconds": 60.35796221913514,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374730.91306057083,
-    "upload_bytes_per_second": 4749748.203097945,
+    "generation_bytes_per_second": 376154.4361440377,
+    "upload_bytes_per_second": 4765332.683014897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.4085306535992,
+    "estimated_full_shards_per_hour": 80.7139855693898,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 5828365,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.2113412039998366,
+      "bytes": 6839007,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -101,3 +101,4 @@
 {"step": 730100, "loss": 0.5311706066131592, "lr": 5e-05, "grad_norm": 0.21856885533574436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106866090687, "tokens_per_second": 135082.6356285108, "data_wait_s": 1.7092999769374728e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 730200, "loss": 0.5325618982315063, "lr": 5e-05, "grad_norm": 0.2232683014262911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106880724396, "tokens_per_second": 151562.0798748091, "data_wait_s": 1.231299938808661e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 730300, "loss": 0.5685901641845703, "lr": 5e-05, "grad_norm": 0.2754684074547258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106895188806, "tokens_per_second": 150372.91048703482, "data_wait_s": 8.905999493435957e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 730400, "loss": 0.553598165512085, "lr": 5e-05, "grad_norm": 0.22882784487219351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106909731207, "tokens_per_second": 150398.9126238805, "data_wait_s": 1.2379998224787414e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a75fa7a20702240f72eedfc43c35740953e48cd7af019f7c7ab96865f5ba4e3e
-size 219885
+oid sha256:bfbff6f9537fc1c27e575c4181f95220db513662080d8ec34d6c6fae0a34e860
+size 222003