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": 76111.2788490324,
-      "required_pages_per_second_upper_estimate": 0.006597132748361484
+      "required_token_positions_per_second": 74459.49814495168,
+      "required_pages_per_second_upper_estimate": 0.0064539605833316875
     },
     {
       "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": 9513.90985612905,
-      "required_pages_per_second_upper_estimate": 0.003795240887238332
+      "required_token_positions_per_second": 9307.43726811896,
+      "required_pages_per_second_upper_estimate": 0.003712875884840817
     },
     {
       "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": 19027.8197122581,
-      "required_pages_per_second_upper_estimate": 0.004401751212241486
+      "required_token_positions_per_second": 18614.87453623792,
+      "required_pages_per_second_upper_estimate": 0.0043062236133560965
     },
     {
       "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": 28541.729568387153,
-      "required_pages_per_second_upper_estimate": 0.001738041583832781
+      "required_token_positions_per_second": 27922.31180435688,
+      "required_pages_per_second_upper_estimate": 0.0017003222918371848
     },
     {
       "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": 28541.729568387153,
-      "required_pages_per_second_upper_estimate": 0.0021972213507386613
+      "required_token_positions_per_second": 27922.31180435688,
+      "required_pages_per_second_upper_estimate": 0.002149536856605497
     },
     {
       "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": 28541.729568387153,
-      "required_pages_per_second_upper_estimate": 0.35439275821531907
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27922.31180435688,
+      "required_pages_per_second_upper_estimate": 0.3467016626439634
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11296909990585391,
-    "producer_p95_seconds": 0.3578753840047284,
-    "gpu_update_seconds": 0.9686869162485147,
-    "producer_batches_per_second": 8.8519781146648,
-    "gpu_batches_per_second": 1.0323252882084473,
-    "producer_headroom": 8.574795382593985,
+    "producer_batch_seconds": 0.22331269892214323,
+    "producer_p95_seconds": 0.37274333600362297,
+    "gpu_update_seconds": 0.990175892086626,
+    "producer_batches_per_second": 4.478025678014149,
+    "gpu_batches_per_second": 1.0099215785719358,
+    "producer_headroom": 4.434033070514479,
     "sustainable": true,
-    "buffer_seconds": 60.05858880740791,
+    "buffer_seconds": 60.40072941728419,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420946.43669003324,
+    "generation_bytes_per_second": 459309.49513353594,
     "upload_bytes_per_second": 5096750.84439948,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.3253061821532,
+    "estimated_full_shards_per_hour": 98.5571254778343,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -369,3 +369,4 @@
 {"step": 191900, "loss": 0.6047444399446249, "lr": 0.0003, "grad_norm": 0.1811759025553319, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28086780349, "tokens_per_second": 152583.00853259044, "data_wait_s": 1.09329994302243e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 192000, "loss": 0.5808647833764553, "lr": 0.0003, "grad_norm": 0.13895726798050276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28101456290, "tokens_per_second": 149713.7505768038, "data_wait_s": 8.994000381790102e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 192100, "loss": 0.5550242215394974, "lr": 0.0003, "grad_norm": 0.14290607580178744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28116035206, "tokens_per_second": 149253.23649242334, "data_wait_s": 7.704002200625837e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 192200, "loss": 0.6105170175433159, "lr": 0.0003, "grad_norm": 0.13729684677486925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28130532083, "tokens_per_second": 147980.23283318375, "data_wait_s": 1.3123004464432597e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6f033b60a5df60a0fd7aed009c18df9d2049c92d24d25b04f7f9a7e515841cf
-size 791424
+oid sha256:e9e32eb2cc5e12c6803f5b860b20d5ea3d97689ecb91aabf2de20f7b73610703
+size 793542