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": 77157.58448488155,
-      "required_pages_per_second_upper_estimate": 0.005337940724925667
+      "required_token_positions_per_second": 77213.31436002062,
+      "required_pages_per_second_upper_estimate": 0.005341796246998914
     },
     {
       "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": 9644.698060610193,
-      "required_pages_per_second_upper_estimate": 0.004038033673779548
+      "required_token_positions_per_second": 9651.664295002578,
+      "required_pages_per_second_upper_estimate": 0.0040409502906481225
     },
     {
       "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": 19289.396121220387,
-      "required_pages_per_second_upper_estimate": 0.004017472516466194
+      "required_token_positions_per_second": 19303.328590005156,
+      "required_pages_per_second_upper_estimate": 0.004020374282290151
     },
     {
       "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": 28934.094181830584,
-      "required_pages_per_second_upper_estimate": 0.0017345547663970215
+      "required_token_positions_per_second": 28954.992885007738,
+      "required_pages_per_second_upper_estimate": 0.0017358076117420197
     },
     {
       "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": 28934.094181830584,
-      "required_pages_per_second_upper_estimate": 0.0022494985550070797
+      "required_token_positions_per_second": 28954.992885007738,
+      "required_pages_per_second_upper_estimate": 0.002251123337255423
     },
     {
       "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": 28934.094181830584,
-      "required_pages_per_second_upper_estimate": 0.3711354931546617
+      "required_token_positions_per_second": 28954.992885007738,
+      "required_pages_per_second_upper_estimate": 0.37140355927973906
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.17026147994579333,
-    "producer_p95_seconds": 0.31315542200172786,
-    "gpu_update_seconds": 0.9555509091195882,
-    "producer_batches_per_second": 5.873319087314248,
-    "gpu_batches_per_second": 1.0465167166460714,
-    "producer_headroom": 5.61225539343256,
+    "producer_batch_seconds": 0.12663615715649712,
+    "producer_p95_seconds": 0.3267651660062256,
+    "gpu_update_seconds": 0.9548612258273264,
+    "producer_batches_per_second": 7.896638862502743,
+    "gpu_batches_per_second": 1.0472726014542728,
+    "producer_headroom": 7.540194264165073,
     "sustainable": true,
-    "buffer_seconds": 60.199707274534056,
+    "buffer_seconds": 60.15625722712156,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311259.59618820884,
-    "upload_bytes_per_second": 5426797.245959924,
+    "generation_bytes_per_second": 310822.1928005138,
+    "upload_bytes_per_second": 5489005.749410033,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.78906358942699,
+    "estimated_full_shards_per_hour": 66.69520700465738,
     "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": 1.3757806130015524,
-      "bytes": 5911161,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0065856220026035,
       "bytes": 6244496,
@@ -562,6 +556,12 @@
       "bytes": 888116,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.3053653089955333,
+      "bytes": 10336652,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -330,3 +330,4 @@
 {"step": 473200, "loss": 0.5413728952407837, "lr": 0.0003, "grad_norm": 0.17737936939567697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69261949976, "tokens_per_second": 150620.79830957143, "data_wait_s": 1.229200279340148e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 473300, "loss": 0.5723682641983032, "lr": 0.0003, "grad_norm": 0.15079825006775918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69276471092, "tokens_per_second": 151624.62627994956, "data_wait_s": 1.9396000425331295e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 473400, "loss": 0.6197728514671326, "lr": 0.0003, "grad_norm": 0.1419205834011256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69291051761, "tokens_per_second": 151552.33377451796, "data_wait_s": 1.4473000192083418e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 473500, "loss": 0.6271270513534546, "lr": 0.0003, "grad_norm": 0.16629573102673484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69305679802, "tokens_per_second": 152218.25866785712, "data_wait_s": 1.1481002729851753e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe3c7fc8ce209c3410e9ac6ab770a230b33affd5d0f60def4f418d104d3ac39c
-size 708039
+oid sha256:eb8ccd3d600bf5b16a05ea3014dbed3ef6014b23bf6a79e61729420afb2dff1e
+size 710157