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": 75361.94889761742,
-      "required_pages_per_second_upper_estimate": 0.006532182727332232
+      "required_token_positions_per_second": 74925.99927086868,
+      "required_pages_per_second_upper_estimate": 0.006494395718589879
     },
     {
       "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": 9420.243612202177,
-      "required_pages_per_second_upper_estimate": 0.003757876022100757
+      "required_token_positions_per_second": 9365.749908858585,
+      "required_pages_per_second_upper_estimate": 0.0037361376690835265
     },
     {
       "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": 18840.487224404354,
-      "required_pages_per_second_upper_estimate": 0.004358415138115704
+      "required_token_positions_per_second": 18731.49981771717,
+      "required_pages_per_second_upper_estimate": 0.00433320281969147
     },
     {
       "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": 28260.730836606534,
-      "required_pages_per_second_upper_estimate": 0.0017209302353537483
+      "required_token_positions_per_second": 28097.249726575752,
+      "required_pages_per_second_upper_estimate": 0.00171097509347197
     },
     {
       "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": 28260.730836606534,
-      "required_pages_per_second_upper_estimate": 0.0021755892905119148
+      "required_token_positions_per_second": 28097.249726575752,
+      "required_pages_per_second_upper_estimate": 0.0021630040621170685
     },
     {
       "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": 28260.730836606534,
-      "required_pages_per_second_upper_estimate": 0.35090369440886215
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28097.249726575752,
+      "required_pages_per_second_upper_estimate": 0.3488738061583589
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.13065600602368477,
-    "producer_p95_seconds": 0.38547299899801146,
-    "gpu_update_seconds": 0.9783186485816971,
-    "producer_batches_per_second": 7.653685662324044,
-    "gpu_batches_per_second": 1.0221618502823544,
-    "producer_headroom": 7.48774341383397,
+    "producer_batch_seconds": 0.11996565644511747,
+    "producer_p95_seconds": 0.3884477519968641,
+    "gpu_update_seconds": 0.9840109003212927,
+    "producer_batches_per_second": 8.335718985187109,
+    "gpu_batches_per_second": 1.016248905041079,
+    "producer_headroom": 8.20243834343926,
     "sustainable": true,
-    "buffer_seconds": 60.65575621206522,
+    "buffer_seconds": 60.024664919598855,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 482680.42225628474,
+    "generation_bytes_per_second": 483300.5291500894,
     "upload_bytes_per_second": 5670226.602968239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.57198239103705,
+    "estimated_full_shards_per_hour": 103.70504289509785,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -399,3 +399,4 @@
 {"step": 394900, "loss": 0.6093955338001251, "lr": 0.0003, "grad_norm": 0.16123570419690397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57802033460, "tokens_per_second": 149377.65332230096, "data_wait_s": 1.1211996024940163e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 395000, "loss": 0.5624208748340607, "lr": 0.0003, "grad_norm": 0.14423838606737646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57816763361, "tokens_per_second": 150025.41638299677, "data_wait_s": 1.1039002856705338e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 395100, "loss": 0.5813555866479874, "lr": 0.0003, "grad_norm": 0.15048847348970862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57831475720, "tokens_per_second": 132896.38789515386, "data_wait_s": 1.3146003766451031e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 395200, "loss": 0.5724810808897018, "lr": 0.0003, "grad_norm": 0.1682321396114497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57846131043, "tokens_per_second": 149427.95350579947, "data_wait_s": 9.98400355456397e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7bee6db1da5f4be9ab79dfdbc7e98decc65e24bf2c7a8568f054f2ed0ea2b50b
-size 855747
+oid sha256:9c50b81fa10315efbc0e1e907f26d9da84b388f75d12be8a25ca39cb87afc172
+size 857865