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": 77545.74574441013,
-      "required_pages_per_second_upper_estimate": 0.005364794621518063
+      "required_token_positions_per_second": 77364.14604702007,
+      "required_pages_per_second_upper_estimate": 0.005352231133083267
     },
     {
       "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": 9693.218218051266,
-      "required_pages_per_second_upper_estimate": 0.00405834805048402
+      "required_token_positions_per_second": 9670.518255877509,
+      "required_pages_per_second_upper_estimate": 0.004048844050350982
     },
     {
       "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": 19386.436436102533,
-      "required_pages_per_second_upper_estimate": 0.004037683454930947
+      "required_token_positions_per_second": 19341.036511755017,
+      "required_pages_per_second_upper_estimate": 0.004028227847965879
     },
     {
       "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": 29079.6546541538,
-      "required_pages_per_second_upper_estimate": 0.0017432808944548262
+      "required_token_positions_per_second": 29011.554767632526,
+      "required_pages_per_second_upper_estimate": 0.0017391984102403818
     },
     {
       "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": 29079.6546541538,
-      "required_pages_per_second_upper_estimate": 0.002260815241477356
+      "required_token_positions_per_second": 29011.554767632526,
+      "required_pages_per_second_upper_estimate": 0.0022555207748400715
     },
     {
       "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": 29079.6546541538,
-      "required_pages_per_second_upper_estimate": 0.3730025866029656
+      "required_token_positions_per_second": 29011.554767632526,
+      "required_pages_per_second_upper_estimate": 0.37212907437863196
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07883097167965047,
-    "producer_p95_seconds": 0.21054152699980477,
-    "gpu_update_seconds": 0.950767824749621,
-    "producer_batches_per_second": 12.685369451790498,
-    "gpu_batches_per_second": 1.0517814906739655,
-    "producer_headroom": 12.060841119824143,
+    "producer_batch_seconds": 0.41469655451565757,
+    "producer_p95_seconds": 0.19895376800013764,
+    "gpu_update_seconds": 0.9529995969346045,
+    "producer_batches_per_second": 2.4114017565637704,
+    "gpu_batches_per_second": 1.0493183871394867,
+    "producer_headroom": 2.2980649020526704,
     "sustainable": true,
-    "buffer_seconds": 60.84914078397574,
+    "buffer_seconds": 60.03897460688008,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359017.1829451649,
+    "generation_bytes_per_second": 314015.8654652145,
     "upload_bytes_per_second": 4252136.214812416,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.03672996774874,
+    "estimated_full_shards_per_hour": 67.38049481360746,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -144,3 +144,4 @@
 {"step": 699400, "loss": 0.5447697043418884, "lr": 8e-05, "grad_norm": 0.20386226727672935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102372998603, "tokens_per_second": 151799.83233677337, "data_wait_s": 1.4466000720858574e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 699500, "loss": 0.5255951285362244, "lr": 8e-05, "grad_norm": 0.20401954144534976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102387638859, "tokens_per_second": 153350.35853250782, "data_wait_s": 1.0024999937741086e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 699600, "loss": 0.5263522863388062, "lr": 8e-05, "grad_norm": 0.2090223796641753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102402160775, "tokens_per_second": 151392.39442331032, "data_wait_s": 1.4368999472935684e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 699700, "loss": 0.5336575508117676, "lr": 8e-05, "grad_norm": 0.20244858323315756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102416811239, "tokens_per_second": 152592.7532016044, "data_wait_s": 1.0017000022344291e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f83215ef1264155b7e14c80e52db8791bbacdee174dc7e29c26679e0f41c6be
-size 310959
+oid sha256:d427bde785fb3eeacf54cec8e3d7d8ea5b44a3f0620c26db4198db3f80f0f962
+size 313077