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": 76465.24646966028,
-      "required_pages_per_second_upper_estimate": 0.005290043174587139
+      "required_token_positions_per_second": 77369.85742027898,
+      "required_pages_per_second_upper_estimate": 0.00535262625913753
     },
     {
       "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": 9558.155808707535,
-      "required_pages_per_second_upper_estimate": 0.004001800240115629
+      "required_token_positions_per_second": 9671.232177534872,
+      "required_pages_per_second_upper_estimate": 0.004049142954440541
     },
     {
       "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": 19116.31161741507,
-      "required_pages_per_second_upper_estimate": 0.003981423578868864
+      "required_token_positions_per_second": 19342.464355069744,
+      "required_pages_per_second_upper_estimate": 0.0040285252300735755
     },
     {
       "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": 28674.467426122606,
-      "required_pages_per_second_upper_estimate": 0.0017189905388194304
+      "required_token_positions_per_second": 29013.696532604612,
+      "required_pages_per_second_upper_estimate": 0.0017393268057801726
     },
     {
       "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": 28674.467426122606,
-      "required_pages_per_second_upper_estimate": 0.002229313716728195
+      "required_token_positions_per_second": 29013.696532604612,
+      "required_pages_per_second_upper_estimate": 0.0022556872876460226
     },
     {
       "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": 28674.467426122606,
-      "required_pages_per_second_upper_estimate": 0.3678052798979311
+      "required_token_positions_per_second": 29013.696532604612,
+      "required_pages_per_second_upper_estimate": 0.3721565466400458
     }
   ],
   "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": 0.11692141062491146,
+    "producer_batch_seconds": 0.15456262191401038,
     "producer_p95_seconds": 0.3450079829999595,
-    "gpu_update_seconds": 0.9642027378967991,
-    "producer_batches_per_second": 8.552753466241011,
-    "gpu_batches_per_second": 1.0371262813267725,
-    "producer_headroom": 8.24658830870592,
+    "gpu_update_seconds": 0.9529292473618489,
+    "producer_batches_per_second": 6.469869542950311,
+    "gpu_batches_per_second": 1.0493958525970997,
+    "producer_headroom": 6.165327914092989,
     "sustainable": true,
-    "buffer_seconds": 60.744772487498345,
+    "buffer_seconds": 60.03454258379648,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379227.1231856726,
-    "upload_bytes_per_second": 5258525.576411283,
+    "generation_bytes_per_second": 378740.93192160403,
+    "upload_bytes_per_second": 5224843.831335754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.37331268003115,
+    "estimated_full_shards_per_hour": 81.2689873527154,
     "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": 1.1604952449997654,
-      "bytes": 5793744,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8524275599993416,
       "bytes": 1090009,
@@ -562,6 +556,12 @@
       "bytes": 6199597,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.057659579004394,
+      "bytes": 2470177,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -434,3 +434,4 @@
 {"step": 600500, "loss": 0.579653263092041, "lr": 0.0003, "grad_norm": 0.18576084141538043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87895218915, "tokens_per_second": 150383.30023103737, "data_wait_s": 1.3119002687744796e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600600, "loss": 0.5803122520446777, "lr": 0.0003, "grad_norm": 0.17307222379511894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87909912683, "tokens_per_second": 152657.14294403233, "data_wait_s": 1.0349998774472624e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600700, "loss": 0.5804871320724487, "lr": 0.0003, "grad_norm": 0.1952816401444965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87924785186, "tokens_per_second": 154255.86697769235, "data_wait_s": 1.0513998859096318e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 600800, "loss": 0.587449312210083, "lr": 0.0003, "grad_norm": 0.14600904392049197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87939498864, "tokens_per_second": 153686.30472890576, "data_wait_s": 9.312003385275602e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b25777fd7f3660261abe848a8bc449953d59a7127b9747dc7e2b9ce3660b412
-size 930660
+oid sha256:c8c9203b16bc0987eb248158bc4340846933ee7c5b942bb8c6402f44d0fd5949
+size 932778