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": 76803.17404064437,
-      "required_pages_per_second_upper_estimate": 0.0053134217880478115
+      "required_token_positions_per_second": 76779.23632129579,
+      "required_pages_per_second_upper_estimate": 0.005311765721080115
     },
     {
       "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": 9600.396755080546,
-      "required_pages_per_second_upper_estimate": 0.004019485642270742
+      "required_token_positions_per_second": 9597.404540161973,
+      "required_pages_per_second_upper_estimate": 0.004018232864368889
     },
     {
       "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": 19200.793510161093,
-      "required_pages_per_second_upper_estimate": 0.003999018929190527
+      "required_token_positions_per_second": 19194.809080323947,
+      "required_pages_per_second_upper_estimate": 0.0039977725302754764
     },
     {
       "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": 28801.19026524164,
-      "required_pages_per_second_upper_estimate": 0.0017265873795300447
+      "required_token_positions_per_second": 28792.21362048592,
+      "required_pages_per_second_upper_estimate": 0.0017260492433834823
     },
     {
       "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": 28801.19026524164,
-      "required_pages_per_second_upper_estimate": 0.0022391658600748377
+      "required_token_positions_per_second": 28792.21362048592,
+      "required_pages_per_second_upper_estimate": 0.002238467965429168
     },
     {
       "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": 28801.19026524164,
-      "required_pages_per_second_upper_estimate": 0.3694307444137664
+      "required_token_positions_per_second": 28792.21362048592,
+      "required_pages_per_second_upper_estimate": 0.3693156016532102
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08154234182038067,
-    "producer_p95_seconds": 0.2968129239998234,
-    "gpu_update_seconds": 0.959960326131613,
-    "producer_batches_per_second": 12.263567340300009,
-    "gpu_batches_per_second": 1.0417097173481495,
-    "producer_headroom": 11.772538103531394,
+    "producer_batch_seconds": 0.09849773232033954,
+    "producer_p95_seconds": 0.2913139949996548,
+    "gpu_update_seconds": 0.9602596161737352,
+    "producer_batches_per_second": 10.152517996533637,
+    "gpu_batches_per_second": 1.0413850412502141,
+    "producer_headroom": 9.749053034548329,
     "sustainable": true,
-    "buffer_seconds": 60.477500546291616,
+    "buffer_seconds": 60.49635581894532,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345666.0653358303,
-    "upload_bytes_per_second": 4841759.838743776,
+    "generation_bytes_per_second": 399038.531660432,
+    "upload_bytes_per_second": 4770673.235171047,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.17189092689688,
+    "estimated_full_shards_per_hour": 85.62437975272866,
     "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": 1.483174205000978,
-      "bytes": 5900678,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8595360259969311,
       "bytes": 1182793,
@@ -562,6 +556,12 @@
       "bytes": 13297372,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.798427575002279,
+      "bytes": 1107567,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -282,3 +282,4 @@
 {"step": 870700, "loss": 0.520969808101654, "lr": 3e-05, "grad_norm": 0.30597625598022377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127446027794, "tokens_per_second": 150416.08419461123, "data_wait_s": 1.535900082672015e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 870800, "loss": 0.49362891912460327, "lr": 3e-05, "grad_norm": 0.30498236498352627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127460642191, "tokens_per_second": 151942.53239090004, "data_wait_s": 1.5082998288562521e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 870900, "loss": 0.5143728256225586, "lr": 3e-05, "grad_norm": 0.29668275921008563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127475238859, "tokens_per_second": 151386.29090419709, "data_wait_s": 1.0212999768555164e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 871000, "loss": 0.5876473188400269, "lr": 3e-05, "grad_norm": 0.3304013484529452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127489898616, "tokens_per_second": 153219.22070710637, "data_wait_s": 1.2127999070798978e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:263621ed126a3623d1878b7668ab5ef1ea7955996740eb6d2cba5b2cb5022c28
-size 606375
+oid sha256:7c918243aadddc9133ea9cd72c9ed9b90a8995f6b9add32a6dac8257447022be
+size 608493