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": 76056.08988933504,
-      "required_pages_per_second_upper_estimate": 0.006592349109210056
+      "required_token_positions_per_second": 75964.18257981725,
+      "required_pages_per_second_upper_estimate": 0.006584382816558001
     },
     {
       "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": 9507.01123616688,
-      "required_pages_per_second_upper_estimate": 0.003792488924591862
+      "required_token_positions_per_second": 9495.522822477156,
+      "required_pages_per_second_upper_estimate": 0.0037879060246039397
     },
     {
       "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": 19014.02247233376,
-      "required_pages_per_second_upper_estimate": 0.004398559463608119
+      "required_token_positions_per_second": 18991.045644954313,
+      "required_pages_per_second_upper_estimate": 0.004393244178972235
     },
     {
       "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": 28521.03370850064,
-      "required_pages_per_second_upper_estimate": 0.0017367813145484775
+      "required_token_positions_per_second": 28486.56846743147,
+      "required_pages_per_second_upper_estimate": 0.00173468256219251
     },
     {
       "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": 28521.03370850064,
-      "required_pages_per_second_upper_estimate": 0.002195628126154791
+      "required_token_positions_per_second": 28486.56846743147,
+      "required_pages_per_second_upper_estimate": 0.0021929748964913923
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28521.03370850064,
-      "required_pages_per_second_upper_estimate": 0.354135784900116
+      "required_token_positions_per_second": 28486.56846743147,
+      "required_pages_per_second_upper_estimate": 0.35370784195377863
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1165799614453249,
-    "producer_p95_seconds": 0.29833251500167535,
-    "gpu_update_seconds": 0.9693898293651105,
-    "producer_batches_per_second": 8.577803488715274,
-    "gpu_batches_per_second": 1.0315767400354687,
-    "producer_headroom": 8.315235460253149,
+    "producer_batch_seconds": 0.08095315363277678,
+    "producer_p95_seconds": 0.22668197999882977,
+    "gpu_update_seconds": 0.9705626717240373,
+    "producer_batches_per_second": 12.352823270310674,
+    "gpu_batches_per_second": 1.0303301673694831,
+    "producer_headroom": 11.989189156567587,
     "sustainable": true,
-    "buffer_seconds": 60.102169420636855,
+    "buffer_seconds": 60.17488564689032,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325639.750221462,
-    "upload_bytes_per_second": 5838899.960048449,
+    "generation_bytes_per_second": 340330.0529882425,
+    "upload_bytes_per_second": 5832080.390755067,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.87470989210982,
+    "estimated_full_shards_per_hour": 73.02690689311463,
     "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": 0.9249057269989862,
-      "bytes": 914039,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 1.1632707989992923,
       "bytes": 10884064,
@@ -562,6 +556,12 @@
       "bytes": 7426505,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.038269829001365,
+      "bytes": 1088045,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -173,3 +173,4 @@
 {"step": 132300, "loss": 0.6078644264489412, "lr": 0.0003, "grad_norm": 0.13996294146397065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19361387048, "tokens_per_second": 151207.65245956017, "data_wait_s": 1.1065996659453958e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 132400, "loss": 0.5686908140778542, "lr": 0.0003, "grad_norm": 0.14668115245836916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19375992526, "tokens_per_second": 149281.23510127407, "data_wait_s": 1.4981000276748091e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 132500, "loss": 0.5837138537317514, "lr": 0.0003, "grad_norm": 0.15276839159771827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19390474475, "tokens_per_second": 147446.5283283952, "data_wait_s": 1.0273000953020528e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 132600, "loss": 0.6222973987460136, "lr": 0.0003, "grad_norm": 0.14609830944624297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19405003107, "tokens_per_second": 149007.31623234323, "data_wait_s": 1.2625001545529813e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ed3eef9dfe2362378a21b1bf42dfee0edcfd5c095dd56dc08bc8b853c12fde75
-size 373164
+oid sha256:7cbd01c3892a4564c4435cbf0c35bf3df93ef376f3618335509d855a8e00e13b
+size 375282