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": 76482.7003063924,
-      "required_pages_per_second_upper_estimate": 0.005291250671510749
+      "required_token_positions_per_second": 76700.75620145099,
+      "required_pages_per_second_upper_estimate": 0.005306336284290286
     },
     {
       "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": 9560.33753829905,
-      "required_pages_per_second_upper_estimate": 0.004002713684735901
+      "required_token_positions_per_second": 9587.594525181374,
+      "required_pages_per_second_upper_estimate": 0.004014125615952919
     },
     {
       "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": 19120.6750765981,
-      "required_pages_per_second_upper_estimate": 0.003982332372344533
+      "required_token_positions_per_second": 19175.189050362747,
+      "required_pages_per_second_upper_estimate": 0.0039936861954495435
     },
     {
       "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": 28681.01261489715,
-      "required_pages_per_second_upper_estimate": 0.0017193829128925395
+      "required_token_positions_per_second": 28762.78357554412,
+      "required_pages_per_second_upper_estimate": 0.001724284957126298
     },
     {
       "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": 28681.01261489715,
-      "required_pages_per_second_upper_estimate": 0.002229822576366173
+      "required_token_positions_per_second": 28762.78357554412,
+      "required_pages_per_second_upper_estimate": 0.00223617990888403
     },
     {
       "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": 28681.01261489715,
-      "required_pages_per_second_upper_estimate": 0.3678892345518548
+      "required_token_positions_per_second": 28762.78357554412,
+      "required_pages_per_second_upper_estimate": 0.36893810463621707
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.12283633717180464,
-    "producer_p95_seconds": 0.34663230599835515,
-    "gpu_update_seconds": 0.9639827007237326,
-    "producer_batches_per_second": 8.140913536043925,
-    "gpu_batches_per_second": 1.0373630141383519,
-    "producer_headroom": 7.847699816834015,
+    "producer_batch_seconds": 0.09719443544497608,
+    "producer_p95_seconds": 0.35123506800300675,
+    "gpu_update_seconds": 0.9612421526374111,
+    "producer_batches_per_second": 10.288654853766007,
+    "gpu_batches_per_second": 1.040320586499715,
+    "producer_headroom": 9.889888739377385,
     "sustainable": true,
-    "buffer_seconds": 60.73091014559515,
+    "buffer_seconds": 60.5582556161569,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340263.87356194935,
-    "upload_bytes_per_second": 5128029.592923862,
+    "generation_bytes_per_second": 339303.67716734187,
+    "upload_bytes_per_second": 5073401.290752656,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.01270632881031,
+    "estimated_full_shards_per_hour": 72.80667053475563,
     "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": 1.3052810369990766,
-      "bytes": 12708434,
-      "pages": 30,
-      "shards": 4
-    },
     {
       "seconds": 1.1760405560016807,
       "bytes": 6277751,
@@ -562,6 +556,12 @@
       "bytes": 1106234,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 0.968032641998434,
+      "bytes": 6353702,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -401,3 +401,4 @@
 {"step": 437200, "loss": 0.6126549243927002, "lr": 0.0003, "grad_norm": 0.16073425316825568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63993467293, "tokens_per_second": 150883.4941976456, "data_wait_s": 1.3878001482225955e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 437300, "loss": 0.5565177202224731, "lr": 0.0003, "grad_norm": 0.1600456835460791, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64008223970, "tokens_per_second": 153196.38968937207, "data_wait_s": 1.2307005818001926e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 437400, "loss": 0.6454831957817078, "lr": 0.0003, "grad_norm": 0.15734464689470082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64022942316, "tokens_per_second": 153265.52283383004, "data_wait_s": 1.1652002285700291e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 437500, "loss": 0.567116379737854, "lr": 0.0003, "grad_norm": 0.1992623283552305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64037726064, "tokens_per_second": 153628.012736391, "data_wait_s": 1.5614001313224435e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8fbc51efad8e2f87500226d918301c75a41f3988ebf467446fc97413d0b5bdbf
-size 859983
+oid sha256:fe8d5de78bb7659eae83ea606a6c9783ba1301a71c550e52531e9337bc218521
+size 862101