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": 75516.9638888524,
-      "required_pages_per_second_upper_estimate": 0.006545619033890578
+      "required_token_positions_per_second": 75100.97275233503,
+      "required_pages_per_second_upper_estimate": 0.0065095619764971475
     },
     {
       "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": 9439.62048610655,
-      "required_pages_per_second_upper_estimate": 0.003765605746811294
+      "required_token_positions_per_second": 9387.621594041879,
+      "required_pages_per_second_upper_estimate": 0.0037448626113139776
     },
     {
       "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": 18879.2409722131,
-      "required_pages_per_second_upper_estimate": 0.0043673801356285905
+      "required_token_positions_per_second": 18775.243188083758,
+      "required_pages_per_second_upper_estimate": 0.004343322078568755
     },
     {
       "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": 28318.861458319654,
-      "required_pages_per_second_upper_estimate": 0.0017244700852282757
+      "required_token_positions_per_second": 28162.864782125638,
+      "required_pages_per_second_upper_estimate": 0.001714970706099367
     },
     {
       "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": 28318.861458319654,
-      "required_pages_per_second_upper_estimate": 0.0021800643466872507
+      "required_token_positions_per_second": 28162.864782125638,
+      "required_pages_per_second_upper_estimate": 0.0021680552907273978
     },
     {
       "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": 28318.861458319654,
-      "required_pages_per_second_upper_estimate": 0.35162548217986334
+      "required_token_positions_per_second": 28162.864782125638,
+      "required_pages_per_second_upper_estimate": 0.349688525548824
     }
   ],
   "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": 0.10932132177333642,
-    "producer_p95_seconds": 0.33024373299849685,
-    "gpu_update_seconds": 0.9763104367982081,
-    "producer_batches_per_second": 9.147346407623669,
-    "gpu_batches_per_second": 1.0242643756626033,
-    "producer_headroom": 8.930649766771582,
+    "producer_batch_seconds": 0.141239377453104,
+    "producer_p95_seconds": 0.36312663400167366,
+    "gpu_update_seconds": 0.9817183093371804,
+    "producer_batches_per_second": 7.080178474533648,
+    "gpu_batches_per_second": 1.0186221347701692,
+    "producer_headroom": 6.95074084182467,
     "sustainable": true,
-    "buffer_seconds": 60.5312470814889,
+    "buffer_seconds": 60.86653517890519,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316358.4668935112,
-    "upload_bytes_per_second": 4176371.8546989355,
+    "generation_bytes_per_second": 310116.1313429996,
+    "upload_bytes_per_second": 4194970.162123537,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.88316254714968,
+    "estimated_full_shards_per_hour": 66.54370265214435,
     "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": 1.1477944020007271,
-      "bytes": 6045290,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1455497599999944,
       "bytes": 7253423,
@@ -562,6 +556,12 @@
       "bytes": 6504817,
       "pages": 23,
       "shards": 2
+    },
+    {
+      "seconds": 1.0333191259996966,
+      "bytes": 7473380,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -288,3 +288,4 @@
 {"step": 343800, "loss": 0.5572164366021752, "lr": 0.0003, "grad_norm": 0.1561984124133296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50322878578, "tokens_per_second": 130001.91361225491, "data_wait_s": 1.4461998944170773e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 343900, "loss": 0.6092987880110741, "lr": 0.0003, "grad_norm": 0.1751567525182395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50337466916, "tokens_per_second": 147868.58429155126, "data_wait_s": 1.152600088971667e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 344000, "loss": 0.5531006511300802, "lr": 0.0003, "grad_norm": 0.14453123897165288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50352088987, "tokens_per_second": 148383.51602231924, "data_wait_s": 1.5735000488348305e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 344100, "loss": 0.5920410379767418, "lr": 0.0003, "grad_norm": 0.16994119213599157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50366769807, "tokens_per_second": 149389.57392022002, "data_wait_s": 1.4823999663349241e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:357126791ceb3aedcd945e8bc5618eeefe14329fb47a69c7e2fa380658347ce5
-size 618300
+oid sha256:7a6a3fdd0b9b22afe8fdc7ce3f6377be13ec34a9f792ee9bf160dddbae87eb04
+size 620418