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": 74721.70012893135,
-      "required_pages_per_second_upper_estimate": 0.0064766875867581924
+      "required_token_positions_per_second": 75587.7969737047,
+      "required_pages_per_second_upper_estimate": 0.0065517586661607615
     },
     {
       "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": 9340.212516116419,
-      "required_pages_per_second_upper_estimate": 0.00372595042130063
+      "required_token_positions_per_second": 9448.474621713087,
+      "required_pages_per_second_upper_estimate": 0.003769137793885865
     },
     {
       "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": 18680.425032232837,
-      "required_pages_per_second_upper_estimate": 0.004321387566955193
+      "required_token_positions_per_second": 18896.949243426174,
+      "required_pages_per_second_upper_estimate": 0.004371476632518806
     },
     {
       "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": 28020.637548349256,
-      "required_pages_per_second_upper_estimate": 0.001706309813771011
+      "required_token_positions_per_second": 28345.423865139262,
+      "required_pages_per_second_upper_estimate": 0.001726087596441412
     },
     {
       "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": 28020.637548349256,
-      "required_pages_per_second_upper_estimate": 0.0021571062445611103
+      "required_token_positions_per_second": 28345.423865139262,
+      "required_pages_per_second_upper_estimate": 0.0021821091942936725
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28020.637548349256,
-      "required_pages_per_second_upper_estimate": 0.3479225393092523
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28345.423865139262,
+      "required_pages_per_second_upper_estimate": 0.35195529837390593
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12300491612501219,
-    "producer_p95_seconds": 0.37420265899709193,
-    "gpu_update_seconds": 0.9867013179944149,
-    "producer_batches_per_second": 8.129756366677908,
-    "gpu_batches_per_second": 1.0134779205855489,
-    "producer_headroom": 8.021641321974577,
+    "producer_batch_seconds": 0.12198301791369204,
+    "producer_p95_seconds": 0.37699189400154864,
+    "gpu_update_seconds": 0.9753955393838019,
+    "producer_batches_per_second": 8.197862432847339,
+    "gpu_batches_per_second": 1.0252251108629653,
+    "producer_headroom": 7.996158449481337,
     "sustainable": true,
-    "buffer_seconds": 60.18878039765931,
+    "buffer_seconds": 60.47452344179572,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289058.626354518,
-    "upload_bytes_per_second": 5176381.674066845,
+    "generation_bytes_per_second": 286113.3896229766,
+    "upload_bytes_per_second": 5180145.650872773,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.025252275244284,
+    "estimated_full_shards_per_hour": 61.39327303425763,
     "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": 1.4560243180021644,
-      "bytes": 7277270,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.4158115079990239,
       "bytes": 6180346,
@@ -562,6 +556,12 @@
       "bytes": 5100299,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1184163700017962,
+      "bytes": 5831579,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -400,3 +400,4 @@
 {"step": 277100, "loss": 0.5619482211768627, "lr": 0.0003, "grad_norm": 0.13645759299485968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40559593939, "tokens_per_second": 150571.18316649858, "data_wait_s": 1.503700332250446e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277200, "loss": 0.672737043350935, "lr": 0.0003, "grad_norm": 0.19240088435973923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40574057787, "tokens_per_second": 123632.47096469428, "data_wait_s": 1.122899993788451e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277300, "loss": 0.5807579811662436, "lr": 0.0003, "grad_norm": 0.15799625978100793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40588611599, "tokens_per_second": 147551.60372950922, "data_wait_s": 1.6421996406279504e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 277400, "loss": 0.5671660769730806, "lr": 0.0003, "grad_norm": 0.14708343725716377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40603211126, "tokens_per_second": 148392.47630144472, "data_wait_s": 1.7530997865833342e-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 sha256:5a5a48d0acabc28b595b539621a8d63c998184b5a17e69d8c5ba34f9fe7497ee
-size 857865
+oid sha256:71dc716a1fc8d80138fa865bf696e830831a7e52bae973dbc5c9c12606e319f6
+size 859983