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": 75170.76083983257,
-      "required_pages_per_second_upper_estimate": 0.006515611031045144
+      "required_token_positions_per_second": 74843.32372358786,
+      "required_pages_per_second_upper_estimate": 0.006487229611690837
     },
     {
       "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": 9396.345104979071,
-      "required_pages_per_second_upper_estimate": 0.003748342550254935
+      "required_token_positions_per_second": 9355.415465448483,
+      "required_pages_per_second_upper_estimate": 0.003732015105093539
     },
     {
       "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": 18792.690209958142,
-      "required_pages_per_second_upper_estimate": 0.004347358140022297
+      "required_token_positions_per_second": 18710.830930896966,
+      "required_pages_per_second_upper_estimate": 0.004328421436485599
     },
     {
       "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": 28189.03531493721,
-      "required_pages_per_second_upper_estimate": 0.0017165643542414175
+      "required_token_positions_per_second": 28066.246396345447,
+      "required_pages_per_second_upper_estimate": 0.0017090871533228468
     },
     {
       "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": 28189.03531493721,
-      "required_pages_per_second_upper_estimate": 0.002170069970787904
+      "required_token_positions_per_second": 28066.246396345447,
+      "required_pages_per_second_upper_estimate": 0.002160617339933229
     },
     {
       "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": 28189.03531493721,
-      "required_pages_per_second_upper_estimate": 0.3500134759792047
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28066.246396345447,
+      "required_pages_per_second_upper_estimate": 0.34848884855837003
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 688946,
-    "producer_batch_seconds": 0.12554474412513628,
-    "producer_p95_seconds": 0.3390076570030942,
-    "gpu_update_seconds": 0.9808068878947936,
-    "producer_batches_per_second": 7.965287650777746,
-    "gpu_batches_per_second": 1.0195686962867916,
-    "producer_headroom": 7.812408991946153,
+    "producer_batch_seconds": 0.1013463907891321,
+    "producer_p95_seconds": 0.3441857650032034,
+    "gpu_update_seconds": 0.9850978862495874,
+    "producer_batches_per_second": 9.86714960654756,
+    "gpu_batches_per_second": 1.015127546164115,
+    "producer_headroom": 9.720108220718448,
     "sustainable": true,
-    "buffer_seconds": 60.810027049477206,
+    "buffer_seconds": 60.09097106122483,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435553.9938756718,
-    "upload_bytes_per_second": 5908724.956912391,
+    "generation_bytes_per_second": 418032.50932937633,
+    "upload_bytes_per_second": 5834605.561030524,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.4597479076635,
+    "estimated_full_shards_per_hour": 89.70004520331352,
     "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": 2.029252282998641,
-      "bytes": 16530913,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 0.8704905230006261,
       "bytes": 1186552,
@@ -562,6 +556,12 @@
       "bytes": 6044525,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4008962299994892,
+      "bytes": 7557731,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -271,3 +271,4 @@
 {"step": 27100, "loss": 0.6737250126898289, "lr": 0.0003, "grad_norm": 0.13846469064501848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3965767704, "tokens_per_second": 136907.83867136188, "data_wait_s": 1.0435000149300322e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 27200, "loss": 0.7051130346953869, "lr": 0.0003, "grad_norm": 0.14441053903586124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3980423900, "tokens_per_second": 149263.70003024803, "data_wait_s": 1.046399847837165e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 27300, "loss": 0.6365917362272739, "lr": 0.0003, "grad_norm": 0.13074144504816781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3995103119, "tokens_per_second": 148590.62108850246, "data_wait_s": 1.2453001545509323e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 27400, "loss": 0.6314107235521078, "lr": 0.0003, "grad_norm": 0.13512946290810118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4009838097, "tokens_per_second": 150100.6244749579, "data_wait_s": 1.3975997717352584e-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 sha256:563ebcb9aece5675f49ec28b855d3dea44db87946201dff189cf06177d3ffe75
-size 576520
+oid sha256:839493dfc0f27e7c3f5ddce5a2654d30f01f3bd48f15468c40e479a922ff240b
+size 578638