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": 76022.21408185888,
-      "required_pages_per_second_upper_estimate": 0.006589412840075472
+      "required_token_positions_per_second": 74120.85582841175,
+      "required_pages_per_second_upper_estimate": 0.00642460792561511
     },
     {
       "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": 9502.77676023236,
-      "required_pages_per_second_upper_estimate": 0.003790799728830525
+      "required_token_positions_per_second": 9265.106978551468,
+      "required_pages_per_second_upper_estimate": 0.0036959896994381157
     },
     {
       "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": 19005.55352046472,
-      "required_pages_per_second_upper_estimate": 0.0043966003206416795
+      "required_token_positions_per_second": 18530.213957102937,
+      "required_pages_per_second_upper_estimate": 0.004286638878348522
     },
     {
       "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": 28508.330280697082,
-      "required_pages_per_second_upper_estimate": 0.0017360077424449756
+      "required_token_positions_per_second": 27795.320935654403,
+      "required_pages_per_second_upper_estimate": 0.0016925892142028002
     },
     {
       "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": 28508.330280697082,
-      "required_pages_per_second_upper_estimate": 0.002194650181117134
+      "required_token_positions_per_second": 27795.320935654403,
+      "required_pages_per_second_upper_estimate": 0.002139760747999557
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28508.330280697082,
-      "required_pages_per_second_upper_estimate": 0.35397805084243367
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27795.320935654403,
+      "required_pages_per_second_upper_estimate": 0.345124861065776
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2869732708672039,
-    "producer_p95_seconds": 0.18901343699963036,
-    "gpu_update_seconds": 0.969821793411745,
-    "producer_batches_per_second": 3.4846450924788295,
-    "gpu_batches_per_second": 1.0311172699904905,
-    "producer_headroom": 3.3794847529912544,
+    "producer_batch_seconds": 0.05051487904689367,
+    "producer_p95_seconds": 0.1732883770000626,
+    "gpu_update_seconds": 0.9946997936812657,
+    "producer_batches_per_second": 19.7961475681588,
+    "gpu_batches_per_second": 1.0053284481935185,
+    "producer_headroom": 19.691223901731448,
     "sustainable": true,
-    "buffer_seconds": 60.12895119152819,
+    "buffer_seconds": 60.67668741455721,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 473159.25148700457,
-    "upload_bytes_per_second": 4597171.145710632,
+    "generation_bytes_per_second": 474460.37868264236,
+    "upload_bytes_per_second": 4633580.1739056045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.52896078546145,
+    "estimated_full_shards_per_hour": 101.80815239295438,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 6703440,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0874789810000038,
+      "bytes": 6434937,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -33,3 +33,4 @@
 {"step": 240400, "loss": 0.5970120839774609, "lr": 0.0003, "grad_norm": 0.15777480089589546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35187296028, "tokens_per_second": 149482.5011255308, "data_wait_s": 1.2068999694747617e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 240500, "loss": 0.5818395670503378, "lr": 0.0003, "grad_norm": 0.15575061846174584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35201777894, "tokens_per_second": 147240.0394176374, "data_wait_s": 1.1921999885089463e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 240600, "loss": 0.5745463036000729, "lr": 0.0003, "grad_norm": 0.13907632765030298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35216349372, "tokens_per_second": 147902.87925055713, "data_wait_s": 1.183400036097737e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 240700, "loss": 0.6726024448871613, "lr": 0.0003, "grad_norm": 0.18383537696960006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35231027010, "tokens_per_second": 149761.4065410573, "data_wait_s": 1.1454000741650816e-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:b600514739136da2fcf5eb623f421c257b5fb516d12f3e8e5965ec11dbba81ec
-size 75078
+oid sha256:765bbca556a893f46ec63f0238d72ccc47a393f003adec0d12c480e6b683921e
+size 77196