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": 77040.6869201237,
-      "required_pages_per_second_upper_estimate": 0.005329853480155999
+      "required_token_positions_per_second": 77024.8875533843,
+      "required_pages_per_second_upper_estimate": 0.0053287604433053855
     },
     {
       "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": 9630.085865015462,
-      "required_pages_per_second_upper_estimate": 0.004031915852621376
+      "required_token_positions_per_second": 9628.110944173037,
+      "required_pages_per_second_upper_estimate": 0.004031088994505689
     },
     {
       "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": 19260.171730030925,
-      "required_pages_per_second_upper_estimate": 0.004011385846480452
+      "required_token_positions_per_second": 19256.221888346074,
+      "required_pages_per_second_upper_estimate": 0.004010563198621827
     },
     {
       "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": 28890.257595046387,
-      "required_pages_per_second_upper_estimate": 0.0017319268299539013
+      "required_token_positions_per_second": 28884.33283251911,
+      "required_pages_per_second_upper_estimate": 0.0017315716494869795
     },
     {
       "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": 28890.257595046387,
-      "required_pages_per_second_upper_estimate": 0.0022460904532013762
+      "required_token_positions_per_second": 28884.33283251911,
+      "required_pages_per_second_upper_estimate": 0.0022456298289751566
     },
     {
       "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": 28890.257595046387,
-      "required_pages_per_second_upper_estimate": 0.3705732044874538
+      "required_token_positions_per_second": 28884.33283251911,
+      "required_pages_per_second_upper_estimate": 0.3704972079952683
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.060806938562482316,
-    "producer_p95_seconds": 0.14038429999982327,
-    "gpu_update_seconds": 0.9570008127840511,
-    "producer_batches_per_second": 16.445491643563795,
-    "gpu_batches_per_second": 1.0449311919504625,
-    "producer_headroom": 15.738348869523874,
+    "producer_batch_seconds": 0.04403391468748907,
+    "producer_p95_seconds": 0.13693934799994167,
+    "gpu_update_seconds": 0.9571971130616803,
+    "producer_batches_per_second": 22.709768302388074,
+    "gpu_batches_per_second": 1.0447168993243312,
+    "producer_headroom": 21.73772465734552,
     "sustainable": true,
-    "buffer_seconds": 60.29105120539522,
+    "buffer_seconds": 60.30341812288586,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 69,
+    "upload_pages_per_commit": 72,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369677.18559737474,
-    "upload_bytes_per_second": 5865846.232172906,
+    "generation_bytes_per_second": 369926.95047177444,
+    "upload_bytes_per_second": 5815454.718005266,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.32411838475163,
+    "estimated_full_shards_per_hour": 79.3777121125691,
     "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": [
@@ -244,6 +244,12 @@
       "bytes": 2246420,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.1746671439996135,
+      "bytes": 6226498,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -14,3 +14,4 @@
 {"step": 441600, "loss": 0.5996808409690857, "lr": 0.0003, "grad_norm": 0.15014320042516954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64637254119, "tokens_per_second": 138165.69149354866, "data_wait_s": 1.3468000133798341e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 441700, "loss": 0.5900413393974304, "lr": 0.0003, "grad_norm": 0.15800119962399603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64651752436, "tokens_per_second": 150390.5225978341, "data_wait_s": 1.516900010756217e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 441800, "loss": 0.5788986682891846, "lr": 0.0003, "grad_norm": 0.14750746214914903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64666235958, "tokens_per_second": 150316.80942708254, "data_wait_s": 1.0941999789793044e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 441900, "loss": 0.5938684344291687, "lr": 0.0003, "grad_norm": 0.17759701841207048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64680833791, "tokens_per_second": 151455.08409828198, "data_wait_s": 1.3769000361207873e-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 sha256:9c4da29d5ce64f6b7a5477f461ea7f2869f45be3e768430b9f8cb4fa8796dd06
-size 34053
+oid sha256:67972c0e554a179fe2efbbcf6af0d4816e053430152857ce6c789c9035e8fe0c
+size 36171