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": 75200.51280538672,
-      "required_pages_per_second_upper_estimate": 0.00651818985601371
+      "required_token_positions_per_second": 75480.42371835574,
+      "required_pages_per_second_upper_estimate": 0.006542451824522142
     },
     {
       "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": 9400.06410067334,
-      "required_pages_per_second_upper_estimate": 0.00374982611324132
+      "required_token_positions_per_second": 9435.052964794468,
+      "required_pages_per_second_upper_estimate": 0.0037637836942693743
     },
     {
       "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": 18800.12820134668,
-      "required_pages_per_second_upper_estimate": 0.004349078788425857
+      "required_token_positions_per_second": 18870.105929588935,
+      "required_pages_per_second_upper_estimate": 0.004365266904288749
     },
     {
       "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": 28200.192302020023,
-      "required_pages_per_second_upper_estimate": 0.0017172437562185737
+      "required_token_positions_per_second": 28305.158894383403,
+      "required_pages_per_second_upper_estimate": 0.001723635670976355
     },
     {
       "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": 28200.192302020023,
-      "required_pages_per_second_upper_estimate": 0.002170928866538056
+      "required_token_positions_per_second": 28305.158894383403,
+      "required_pages_per_second_upper_estimate": 0.0021790094853843135
     },
     {
       "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": 28200.192302020023,
-      "required_pages_per_second_upper_estimate": 0.3501520084187395
+      "required_token_positions_per_second": 28305.158894383403,
+      "required_pages_per_second_upper_estimate": 0.35145534219530655
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10071845389842338,
-    "producer_p95_seconds": 0.3278017419979733,
-    "gpu_update_seconds": 0.9804188462225321,
-    "producer_batches_per_second": 9.928667104128907,
-    "gpu_batches_per_second": 1.0199722331459788,
-    "producer_headroom": 9.734252346757671,
+    "producer_batch_seconds": 0.0938382583750581,
+    "producer_p95_seconds": 0.3267776530010451,
+    "gpu_update_seconds": 0.9767830699401655,
+    "producer_batches_per_second": 10.656634269608276,
+    "gpu_batches_per_second": 1.023768767881344,
+    "producer_headroom": 10.409219937097545,
     "sustainable": true,
-    "buffer_seconds": 60.78596846579699,
+    "buffer_seconds": 60.56055033629026,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374111.0763365757,
-    "upload_bytes_per_second": 3439776.7876809277,
+    "generation_bytes_per_second": 409319.4278904717,
+    "upload_bytes_per_second": 3400481.077198823,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.2755281216903,
+    "estimated_full_shards_per_hour": 87.83042075667966,
     "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": 1.1112909670009685,
-      "bytes": 6011206,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9119941530007054,
       "bytes": 5747516,
@@ -562,6 +556,12 @@
       "bytes": 7725078,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0130331639993528,
+      "bytes": 839053,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -271,3 +271,4 @@
 {"step": 67100, "loss": 0.6024973634630442, "lr": 0.0003, "grad_norm": 0.1236716504339559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9820152581, "tokens_per_second": 149987.07305735402, "data_wait_s": 7.930997526273131e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 67200, "loss": 0.6835955232381821, "lr": 0.0003, "grad_norm": 0.14360932361638307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9834836303, "tokens_per_second": 149749.15484064168, "data_wait_s": 9.785999282030389e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 67300, "loss": 0.618693120777607, "lr": 0.0003, "grad_norm": 0.13847754745698787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9849572397, "tokens_per_second": 150969.64528632563, "data_wait_s": 7.422000635415316e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 67400, "loss": 0.6169149242341518, "lr": 0.0003, "grad_norm": 0.12542056418614891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9864109107, "tokens_per_second": 148125.61582019657, "data_wait_s": 9.758001397131011e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:127017911fffa7cce27ed67554d027f624a305f2be8d70783e868f5371701f79
-size 582294
+oid sha256:32530108834e5c29433d5116cbc82ed42a7ffec72c1912aa8b0bd13d51fcaa62
+size 584412