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": 74616.38181288075,
-      "required_pages_per_second_upper_estimate": 0.006467558862049758
+      "required_token_positions_per_second": 74770.46689945106,
+      "required_pages_per_second_upper_estimate": 0.006480914566828591
     },
     {
       "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": 9327.047726610093,
-      "required_pages_per_second_upper_estimate": 0.0037206987899354132
+      "required_token_positions_per_second": 9346.308362431382,
+      "required_pages_per_second_upper_estimate": 0.003728382145536693
     },
     {
       "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": 18654.095453220187,
-      "required_pages_per_second_upper_estimate": 0.004315296682235381
+      "required_token_positions_per_second": 18692.616724862764,
+      "required_pages_per_second_upper_estimate": 0.004324207900478665
     },
     {
       "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": 27981.14317983028,
-      "required_pages_per_second_upper_estimate": 0.0017039048139391429
+      "required_token_positions_per_second": 28038.925087294145,
+      "required_pages_per_second_upper_estimate": 0.001707423428945453
     },
     {
       "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": 27981.14317983028,
-      "required_pages_per_second_upper_estimate": 0.00215406585874511
+      "required_token_positions_per_second": 28038.925087294145,
+      "required_pages_per_second_upper_estimate": 0.0021585140699322356
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27981.14317983028,
-      "required_pages_per_second_upper_estimate": 0.3474321514313953
+      "required_token_positions_per_second": 28038.925087294145,
+      "required_pages_per_second_upper_estimate": 0.3481496093384922
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11913276657806193,
-    "producer_p95_seconds": 0.32301659399854543,
-    "gpu_update_seconds": 0.9880940111099384,
-    "producer_batches_per_second": 8.393996284345066,
-    "gpu_batches_per_second": 1.0120494495019632,
-    "producer_headroom": 8.294057457840436,
+    "producer_batch_seconds": 0.3627634417499763,
+    "producer_p95_seconds": 0.3269775070002652,
+    "gpu_update_seconds": 0.9860577719696076,
+    "producer_batches_per_second": 2.756617356964045,
+    "gpu_batches_per_second": 1.0141393622429884,
+    "producer_headroom": 2.7181839691807146,
     "sustainable": true,
-    "buffer_seconds": 60.273734677706244,
+    "buffer_seconds": 60.149524090146066,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355577.373587176,
-    "upload_bytes_per_second": 5217106.874362177,
+    "generation_bytes_per_second": 339490.2539706306,
+    "upload_bytes_per_second": 5191784.548656985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.29862695418795,
+    "estimated_full_shards_per_hour": 72.8467055734557,
     "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": 0.9532060609999462,
-      "bytes": 7422959,
-      "pages": 29,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2087593340002059,
-      "bytes": 6148456,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.2602108099999896,
       "bytes": 2561145,
@@ -562,6 +550,18 @@
       "bytes": 6094784,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.438195700000506,
+      "bytes": 12388203,
+      "pages": 31,
+      "shards": 4
+    },
+    {
+      "seconds": 0.9514959919997636,
+      "bytes": 336067,
+      "pages": 9,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 209800, "loss": 0.5944355335086584, "lr": 0.0003, "grad_norm": 0.1378234884751291, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30706866861, "tokens_per_second": 147099.06887911726, "data_wait_s": 1.7036998542607762e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209900, "loss": 0.6171294096857309, "lr": 0.0003, "grad_norm": 0.14364944481939032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30721359198, "tokens_per_second": 146652.94399301035, "data_wait_s": 1.898600021377206e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 210000, "loss": 0.5918615683913231, "lr": 0.0003, "grad_norm": 0.17345168380647322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30736062273, "tokens_per_second": 148821.90664461232, "data_wait_s": 1.5919998986646533e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 210100, "loss": 0.5807416588068008, "lr": 0.0003, "grad_norm": 0.1436917269689581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30750746073, "tokens_per_second": 125072.04918072904, "data_wait_s": 1.2976999641978182e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79cced4b5fca24391b7f9843f95dd80d3f412b48640f059818279ca925dcbdce
-size 320214
+oid sha256:220a0b1f625c5de9173777af3efd8a7192c65fe3b511ebbd8b29e6a407eaee76
+size 322332