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": 75427.73566052444,
-      "required_pages_per_second_upper_estimate": 0.006537884957205987
+      "required_token_positions_per_second": 72529.10542517215,
+      "required_pages_per_second_upper_estimate": 0.006286639034916817
     },
     {
       "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": 9428.466957565555,
-      "required_pages_per_second_upper_estimate": 0.003761156437516178
+      "required_token_positions_per_second": 9066.138178146519,
+      "required_pages_per_second_upper_estimate": 0.003616618070107914
     },
     {
       "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": 18856.93391513111,
-      "required_pages_per_second_upper_estimate": 0.004362219790563468
+      "required_token_positions_per_second": 18132.276356293038,
+      "required_pages_per_second_upper_estimate": 0.004194583017863728
     },
     {
       "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": 28285.400872696668,
-      "required_pages_per_second_upper_estimate": 0.0017224325111179612
+      "required_token_positions_per_second": 27198.41453443956,
+      "required_pages_per_second_upper_estimate": 0.0016562407460946709
     },
     {
       "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": 28285.400872696668,
-      "required_pages_per_second_upper_estimate": 0.002177488458181164
+      "required_token_positions_per_second": 27198.41453443956,
+      "required_pages_per_second_upper_estimate": 0.002093809240891365
     },
     {
       "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": 28285.400872696668,
-      "required_pages_per_second_upper_estimate": 0.3512100136918021
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27198.41453443956,
+      "required_pages_per_second_upper_estimate": 0.33771328127990313
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05929656364840241,
-    "producer_p95_seconds": 0.20810847400025523,
-    "gpu_update_seconds": 0.977465376023292,
-    "producer_batches_per_second": 16.864383675409535,
-    "gpu_batches_per_second": 1.023054140360846,
-    "producer_headroom": 16.484351130685248,
+    "producer_batch_seconds": 0.43021664167174833,
+    "producer_p95_seconds": 0.24422349899941764,
+    "gpu_update_seconds": 1.0165298409211008,
+    "producer_batches_per_second": 2.3244103159612117,
+    "gpu_batches_per_second": 0.9837389516218011,
+    "producer_headroom": 2.3628324487194163,
     "sustainable": true,
-    "buffer_seconds": 60.6028533134441,
+    "buffer_seconds": 60.99179045526605,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 246085.66055733646,
+    "generation_bytes_per_second": 226958.61020250482,
     "upload_bytes_per_second": 6513806.664100802,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.80425417461463,
+    "estimated_full_shards_per_hour": 48.70003442341193,
     "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": [

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -114,3 +114,4 @@
 {"step": 11400, "loss": 0.7066514007747173, "lr": 0.0003, "grad_norm": 0.15130700510484968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1667767034, "tokens_per_second": 147853.70048813766, "data_wait_s": 1.3994998880662024e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11500, "loss": 0.6995341554284096, "lr": 0.0003, "grad_norm": 0.15938632569844022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1682329231, "tokens_per_second": 148173.18191002306, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11600, "loss": 0.6731462813913822, "lr": 0.0003, "grad_norm": 0.14444594314923076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1696884254, "tokens_per_second": 147854.69487109123, "data_wait_s": 1.1763999282266013e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 11700, "loss": 0.7302270978689194, "lr": 0.0003, "grad_norm": 0.1457306650160281, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1711427439, "tokens_per_second": 146573.4266490581, "data_wait_s": 1.314000110141933e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3fbd76cc6167ca51bc64c239857c28d3e4da4f7cf94f525aab6cf80c01354b9
-size 243300
+oid sha256:777fd5fbfc2f46cff809c7e6ef1dda5778b0e1a564ffcb2b62d849c3ccb4f17e
+size 245383