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": 76814.02817439828,
-      "required_pages_per_second_upper_estimate": 0.005314172702205967
+      "required_token_positions_per_second": 76248.31117994558,
+      "required_pages_per_second_upper_estimate": 0.005275035087885455
     },
     {
       "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": 9601.753521799785,
-      "required_pages_per_second_upper_estimate": 0.0040200536921635764
+      "required_token_positions_per_second": 9531.038897493198,
+      "required_pages_per_second_upper_estimate": 0.003990446955655684
     },
     {
       "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": 19203.50704359957,
-      "required_pages_per_second_upper_estimate": 0.003999584086645072
+      "required_token_positions_per_second": 19062.077794986395,
+      "required_pages_per_second_upper_estimate": 0.003970128103899048
     },
     {
       "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": 28805.260565399352,
-      "required_pages_per_second_upper_estimate": 0.0017268313878095111
+      "required_token_positions_per_second": 28593.11669247959,
+      "required_pages_per_second_upper_estimate": 0.0017141136865529095
     },
     {
       "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": 28805.260565399352,
-      "required_pages_per_second_upper_estimate": 0.002239482308008742
+      "required_token_positions_per_second": 28593.11669247959,
+      "required_pages_per_second_upper_estimate": 0.0022229890550115117
     },
     {
       "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": 28805.260565399352,
-      "required_pages_per_second_upper_estimate": 0.36948295385384167
+      "required_token_positions_per_second": 28593.11669247959,
+      "required_pages_per_second_upper_estimate": 0.3667618000343709
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08833682442170243,
-    "producer_p95_seconds": 0.2469352219995926,
-    "gpu_update_seconds": 0.959824679843742,
-    "producer_batches_per_second": 11.320307318567384,
-    "gpu_batches_per_second": 1.041856935959178,
-    "producer_headroom": 10.865510347776707,
+    "producer_batch_seconds": 0.07084041854670886,
+    "producer_p95_seconds": 0.26022898399969563,
+    "gpu_update_seconds": 0.9669460065286213,
+    "producer_batches_per_second": 14.11623506064757,
+    "gpu_batches_per_second": 1.0341839081481334,
+    "producer_headroom": 13.649637119112478,
     "sustainable": true,
-    "buffer_seconds": 60.468954830155745,
+    "buffer_seconds": 60.91759841130314,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330707.8441307348,
-    "upload_bytes_per_second": 4907760.828050056,
+    "generation_bytes_per_second": 330283.34720341425,
+    "upload_bytes_per_second": 4920355.588237083,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.96220486585172,
+    "estimated_full_shards_per_hour": 70.87111770703144,
     "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": 1.367829029000859,
-      "bytes": 6101351,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9580737389987917,
       "bytes": 2350217,
@@ -562,6 +556,12 @@
       "bytes": 343455,
       "pages": 9,
       "shards": 1
+    },
+    {
+      "seconds": 1.1079528180016496,
+      "bytes": 5894448,
+      "pages": 27,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 420000, "loss": 0.6040761470794678, "lr": 0.0003, "grad_norm": 0.1615146686793636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61477032401, "tokens_per_second": 145438.4825535189, "data_wait_s": 8.343002264155075e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 420100, "loss": 0.5920270085334778, "lr": 0.0003, "grad_norm": 0.14548719408145425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61491509441, "tokens_per_second": 137382.28344275174, "data_wait_s": 1.4730001566931605e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 420200, "loss": 0.5941126346588135, "lr": 0.0003, "grad_norm": 0.181814914446555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61506083731, "tokens_per_second": 150840.83369612007, "data_wait_s": 1.7055001080734655e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 420300, "loss": 0.5611682534217834, "lr": 0.0003, "grad_norm": 0.14011688720046497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61520691441, "tokens_per_second": 151543.81931361408, "data_wait_s": 8.942999556893483e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d26ef9c7fa34e0c56dd6263838110ab261c361704602128bbe2f21ff6cee2ec
-size 493338
+oid sha256:403c96914ec2e6f5e9dad139bb7adfb43c669fa49f0449733cb272454b9ea53c
+size 495456