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": 75897.83098863548,
-      "required_pages_per_second_upper_estimate": 0.0065786316288009355
+      "required_token_positions_per_second": 75665.69836686787,
+      "required_pages_per_second_upper_estimate": 0.00655851096148087
     },
     {
       "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": 9487.228873579435,
-      "required_pages_per_second_upper_estimate": 0.0037845974443830523
+      "required_token_positions_per_second": 9458.212295858484,
+      "required_pages_per_second_upper_estimate": 0.003773022297693667
     },
     {
       "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": 18974.45774715887,
-      "required_pages_per_second_upper_estimate": 0.004389406860754289
+      "required_token_positions_per_second": 18916.42459171697,
+      "required_pages_per_second_upper_estimate": 0.004375981911591458
     },
     {
       "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": 28461.686620738306,
-      "required_pages_per_second_upper_estimate": 0.001733167388273857
+      "required_token_positions_per_second": 28374.636887575452,
+      "required_pages_per_second_upper_estimate": 0.0017278665162388923
     },
     {
       "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": 28461.686620738306,
-      "required_pages_per_second_upper_estimate": 0.0021910594230555927
+      "required_token_positions_per_second": 28374.636887575452,
+      "required_pages_per_second_upper_estimate": 0.002184358093627632
     },
     {
       "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": 28461.686620738306,
-      "required_pages_per_second_upper_estimate": 0.35339889269203356
+      "required_token_positions_per_second": 28374.636887575452,
+      "required_pages_per_second_upper_estimate": 0.3523180263428667
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.06784109200781785,
-    "producer_p95_seconds": 0.16832244499983062,
-    "gpu_update_seconds": 0.971411159444591,
-    "producer_batches_per_second": 14.740328765414954,
-    "gpu_batches_per_second": 1.029430216317213,
-    "producer_headroom": 14.318919856606197,
+    "producer_batch_seconds": 0.13129274778125932,
+    "producer_p95_seconds": 0.1682472989996313,
+    "gpu_update_seconds": 0.9743913238271736,
+    "producer_batches_per_second": 7.61656692314836,
+    "gpu_batches_per_second": 1.026281716130478,
+    "producer_headroom": 7.421516727264794,
     "sustainable": true,
-    "buffer_seconds": 60.22749188556464,
+    "buffer_seconds": 60.412262077284765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313232.85531853064,
-    "upload_bytes_per_second": 5494031.573970095,
+    "generation_bytes_per_second": 351443.47014520573,
+    "upload_bytes_per_second": 5528607.3444076935,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.21247906367245,
+    "estimated_full_shards_per_hour": 75.41158750788811,
     "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": [
@@ -292,6 +292,12 @@
       "bytes": 6095231,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0922540970004775,
+      "bytes": 6780784,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -47,3 +47,4 @@
 {"step": 359700, "loss": 0.5889822412282228, "lr": 0.0003, "grad_norm": 0.1657767608538852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52650059176, "tokens_per_second": 146595.91290202373, "data_wait_s": 9.807999958866276e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 359800, "loss": 0.5995389968156815, "lr": 0.0003, "grad_norm": 0.15052040963303098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52664547186, "tokens_per_second": 147907.74579080864, "data_wait_s": 1.2000999959127512e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 359900, "loss": 0.589072197675705, "lr": 0.0003, "grad_norm": 0.1709890691884968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52679165005, "tokens_per_second": 148789.96262994417, "data_wait_s": 8.371000149054453e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 360000, "loss": 0.6037846226245165, "lr": 0.0003, "grad_norm": 0.14638077382152287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52693835160, "tokens_per_second": 149850.34213180692, "data_wait_s": 9.698000212665647e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9f9a537dbb76ed80173afa6fd0c9fc69088aaf66d8ec5fb041ada4c2890a44d
-size 103947
+oid sha256:659bc93c0f3c0daadb0479e09f2fb126bb0df3be8c9dc11964626108e86b2301
+size 106065