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": 77001.7747409032,
-      "required_pages_per_second_upper_estimate": 0.00532716144530883
+      "required_token_positions_per_second": 76720.47547070158,
+      "required_pages_per_second_upper_estimate": 0.00530770050909205
     },
     {
       "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": 9625.2218426129,
-      "required_pages_per_second_upper_estimate": 0.004029879388013761
+      "required_token_positions_per_second": 9590.059433837698,
+      "required_pages_per_second_upper_estimate": 0.004015157621734176
     },
     {
       "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": 19250.4436852258,
-      "required_pages_per_second_upper_estimate": 0.004009359751293338
+      "required_token_positions_per_second": 19180.118867675395,
+      "required_pages_per_second_upper_estimate": 0.003994712946387743
     },
     {
       "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": 28875.6655278387,
-      "required_pages_per_second_upper_estimate": 0.001731052057805597
+      "required_token_positions_per_second": 28770.178301513093,
+      "required_pages_per_second_upper_estimate": 0.0017247282596570452
     },
     {
       "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": 28875.6655278387,
-      "required_pages_per_second_upper_estimate": 0.0022449559841596993
+      "required_token_positions_per_second": 28770.178301513093,
+      "required_pages_per_second_upper_estimate": 0.0022367548163021557
     },
     {
       "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": 28875.6655278387,
-      "required_pages_per_second_upper_estimate": 0.3703860331170547
+      "required_token_positions_per_second": 28770.178301513093,
+      "required_pages_per_second_upper_estimate": 0.36903295624110893
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.3373699430389365,
-    "producer_p95_seconds": 0.3439469939985429,
-    "gpu_update_seconds": 0.9574844248471045,
-    "producer_batches_per_second": 2.964105192632967,
-    "gpu_batches_per_second": 1.044403411741851,
-    "producer_headroom": 2.838084555554492,
+    "producer_batch_seconds": 0.41118226897634713,
+    "producer_p95_seconds": 0.3353669310017722,
+    "gpu_update_seconds": 0.960995087004585,
+    "producer_batches_per_second": 2.4320114835922655,
+    "gpu_batches_per_second": 1.0405880462063475,
+    "producer_headroom": 2.3371510872708994,
     "sustainable": true,
-    "buffer_seconds": 60.321518765367586,
+    "buffer_seconds": 60.54269048128886,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386012.2469064946,
-    "upload_bytes_per_second": 4946525.851116882,
+    "generation_bytes_per_second": 353220.6740186299,
+    "upload_bytes_per_second": 5000542.364859272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.82924228092317,
+    "estimated_full_shards_per_hour": 75.79293408793613,
     "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": 1.1207633359990723,
-      "bytes": 2404167,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.4597645020003256,
       "bytes": 6256363,
@@ -562,6 +556,12 @@
       "bytes": 6988535,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.6882793909971952,
+      "bytes": 9959393,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -402,3 +402,4 @@
 {"step": 685200, "loss": 0.5691226124763489, "lr": 0.0001, "grad_norm": 0.2106518709980132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100295657942, "tokens_per_second": 151739.2991392258, "data_wait_s": 1.6899000911507756e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685300, "loss": 0.6044577956199646, "lr": 0.0001, "grad_norm": 0.2075308347433876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100310140405, "tokens_per_second": 150438.1131585047, "data_wait_s": 2.190299710491672e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685400, "loss": 0.5695536136627197, "lr": 0.0001, "grad_norm": 0.1888986239217438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100324882735, "tokens_per_second": 153663.03866914738, "data_wait_s": 1.4646997442469e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 685500, "loss": 0.554060161113739, "lr": 0.0001, "grad_norm": 0.20678600802984906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100339543873, "tokens_per_second": 153235.66768583274, "data_wait_s": 8.188995707314461e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd4afb06b3f60e3496b5cb09dc383616ae1a81548a1ed2d0542d391bb9c9124d
-size 862101
+oid sha256:c7495ed3c3b7504c66256336f37e52b23ff1e525249d0f67a28f2081cfbf8115
+size 864219