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": 75676.60191480261,
-      "required_pages_per_second_upper_estimate": 0.006559456053381063
+      "required_token_positions_per_second": 75491.94320750768,
+      "required_pages_per_second_upper_estimate": 0.0065434503046459536
     },
     {
       "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": 9459.575239350326,
-      "required_pages_per_second_upper_estimate": 0.0037735659962303838
+      "required_token_positions_per_second": 9436.49290093846,
+      "required_pages_per_second_upper_estimate": 0.003764358106326177
     },
     {
       "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": 18919.150478700652,
-      "required_pages_per_second_upper_estimate": 0.004376612497571161
+      "required_token_positions_per_second": 18872.98580187692,
+      "required_pages_per_second_upper_estimate": 0.004365933112058553
     },
     {
       "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": 28378.72571805098,
-      "required_pages_per_second_upper_estimate": 0.0017281155045624165
+      "required_token_positions_per_second": 28309.47870281538,
+      "required_pages_per_second_upper_estimate": 0.0017238987246455792
     },
     {
       "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": 28378.72571805098,
-      "required_pages_per_second_upper_estimate": 0.0021846728631162455
+      "required_token_positions_per_second": 28309.47870281538,
+      "required_pages_per_second_upper_estimate": 0.0021793420361954025
     },
     {
       "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": 28378.72571805098,
-      "required_pages_per_second_upper_estimate": 0.3523687959329374
+      "required_token_positions_per_second": 28309.47870281538,
+      "required_pages_per_second_upper_estimate": 0.3515089797585629
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.47847483936709523,
-    "producer_p95_seconds": 0.3735654810006963,
-    "gpu_update_seconds": 0.974250932712381,
-    "producer_batches_per_second": 2.089974054482686,
-    "gpu_batches_per_second": 1.026429604964228,
-    "producer_headroom": 2.0361591719244334,
+    "producer_batch_seconds": 0.15285167817899037,
+    "producer_p95_seconds": 0.3867400009985431,
+    "gpu_update_seconds": 0.9766340203661329,
+    "producer_batches_per_second": 6.54228996314318,
+    "gpu_batches_per_second": 1.0239250109525238,
+    "producer_headroom": 6.389422949105523,
     "sustainable": true,
-    "buffer_seconds": 60.40355782816762,
+    "buffer_seconds": 60.55130926270024,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325437.51489142334,
-    "upload_bytes_per_second": 5123306.2980195945,
+    "generation_bytes_per_second": 317479.3983608873,
+    "upload_bytes_per_second": 5119391.312149676,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.83131489808106,
+    "estimated_full_shards_per_hour": 68.12368834609951,
     "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": 0.9998172479972709,
-      "bytes": 6024653,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9568293169977551,
       "bytes": 2417450,
@@ -562,6 +556,12 @@
       "bytes": 5170958,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.3704213370001526,
+      "bytes": 7604637,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -385,3 +385,4 @@
 {"step": 275600, "loss": 0.6014276780188084, "lr": 0.0003, "grad_norm": 0.14329618538429756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40339605846, "tokens_per_second": 150646.99400476814, "data_wait_s": 1.5729005099274218e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 275700, "loss": 0.5938325859606266, "lr": 0.0003, "grad_norm": 0.15760388608972853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40354109564, "tokens_per_second": 147712.3905512474, "data_wait_s": 1.0187002771999687e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 275800, "loss": 0.5608959067612886, "lr": 0.0003, "grad_norm": 0.14706332915765138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40368593068, "tokens_per_second": 147240.47465103294, "data_wait_s": 1.006499951472506e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 275900, "loss": 0.5654884856194258, "lr": 0.0003, "grad_norm": 0.1431217299136518, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40383055030, "tokens_per_second": 147731.86951555742, "data_wait_s": 1.160100509878248e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9d24546ddeecbf562951b1772268424ed79cc501a5879273e7c99c009115880
-size 826095
+oid sha256:c8c5a5ff968e5f7ad771110556ef1864cf5a35faa21a0d41ee7911362b109443
+size 828213