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": 77092.6594232072,
-      "required_pages_per_second_upper_estimate": 0.005333449058512137
+      "required_token_positions_per_second": 76840.80428014703,
+      "required_pages_per_second_upper_estimate": 0.00531602513533079
     },
     {
       "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": 9636.5824279009,
-      "required_pages_per_second_upper_estimate": 0.004034635827837849
+      "required_token_positions_per_second": 9605.100535018379,
+      "required_pages_per_second_upper_estimate": 0.004021455016704618
     },
     {
       "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": 19273.1648558018,
-      "required_pages_per_second_upper_estimate": 0.004014091971926756
+      "required_token_positions_per_second": 19210.201070036757,
+      "required_pages_per_second_upper_estimate": 0.00400097827581859
     },
     {
       "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": 28909.747283702698,
-      "required_pages_per_second_upper_estimate": 0.0017330952070298156
+      "required_token_positions_per_second": 28815.301605055138,
+      "required_pages_per_second_upper_estimate": 0.0017274333328050956
     },
     {
       "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": 28909.747283702698,
-      "required_pages_per_second_upper_estimate": 0.002247605690768324
+      "required_token_positions_per_second": 28815.301605055138,
+      "required_pages_per_second_upper_estimate": 0.002240262954676113
     },
     {
       "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": 28909.747283702698,
-      "required_pages_per_second_upper_estimate": 0.3708231972871397
+      "required_token_positions_per_second": 28815.301605055138,
+      "required_pages_per_second_upper_estimate": 0.3696117495293177
     }
   ],
   "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": 0.3317292300234911,
-    "producer_p95_seconds": 0.1993053759997565,
-    "gpu_update_seconds": 0.9563556446439785,
-    "producer_batches_per_second": 3.014506740720997,
-    "gpu_batches_per_second": 1.0456361141385524,
-    "producer_headroom": 2.8829405373058474,
+    "producer_batch_seconds": 0.07585416217968088,
+    "producer_p95_seconds": 0.17503454999950918,
+    "gpu_update_seconds": 0.9594902173486064,
+    "producer_batches_per_second": 13.183192210748203,
+    "gpu_batches_per_second": 1.042220110136543,
+    "producer_headroom": 12.649143959639249,
     "sustainable": true,
-    "buffer_seconds": 60.250405612570646,
+    "buffer_seconds": 60.4478836929622,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356585.1289413755,
-    "upload_bytes_per_second": 4309496.645920998,
+    "generation_bytes_per_second": 357462.0739709278,
+    "upload_bytes_per_second": 4295212.239902175,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.51486779385517,
+    "estimated_full_shards_per_hour": 76.70303978296161,
     "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": [
@@ -496,6 +496,12 @@
       "bytes": 10941174,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9096585289998984,
+      "bytes": 2702720,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -67,3 +67,4 @@
 {"step": 446900, "loss": 0.6321390271186829, "lr": 0.0003, "grad_norm": 0.15205760743964034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65412524072, "tokens_per_second": 152354.3841680319, "data_wait_s": 1.3597000361187384e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 447000, "loss": 0.4998493194580078, "lr": 0.0003, "grad_norm": 0.1414208802714209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65427210081, "tokens_per_second": 152827.77021420054, "data_wait_s": 1.2985000466869678e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 447100, "loss": 0.6052762269973755, "lr": 0.0003, "grad_norm": 0.14999873744594025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65441963550, "tokens_per_second": 153995.48498493223, "data_wait_s": 1.2706000234175008e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 447200, "loss": 0.6125456094741821, "lr": 0.0003, "grad_norm": 0.18039655206486915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65456727001, "tokens_per_second": 153944.95189737467, "data_wait_s": 1.4642999303760007e-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 sha256:93e345c529eed39f493705ff060de848666eeea81e6edc308a0d38552f3cf4e0
-size 147090
+oid sha256:3518d841653e1b60776a0c50791d4e8683876e48e01b433f1daee3eb6db72181
+size 149208