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": 74575.6696782461,
-      "required_pages_per_second_upper_estimate": 0.006464030037403588
+      "required_token_positions_per_second": 74048.36666452323,
+      "required_pages_per_second_upper_estimate": 0.00641832474861136
     },
     {
       "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": 9321.958709780762,
-      "required_pages_per_second_upper_estimate": 0.0037186687050346106
+      "required_token_positions_per_second": 9256.045833065404,
+      "required_pages_per_second_upper_estimate": 0.0036923750730275266
     },
     {
       "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": 18643.917419561523,
-      "required_pages_per_second_upper_estimate": 0.004312942173275723
+      "required_token_positions_per_second": 18512.091666130807,
+      "required_pages_per_second_upper_estimate": 0.004282446605273548
     },
     {
       "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": 27965.876129342287,
-      "required_pages_per_second_upper_estimate": 0.0017029751306644474
+      "required_token_positions_per_second": 27768.137499196208,
+      "required_pages_per_second_upper_estimate": 0.0016909338855429648
     },
     {
       "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": 27965.876129342287,
-      "required_pages_per_second_upper_estimate": 0.0021528905589366437
+      "required_token_positions_per_second": 27768.137499196208,
+      "required_pages_per_second_upper_estimate": 0.0021376680918124367
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27965.876129342287,
-      "required_pages_per_second_upper_estimate": 0.3472425857598655
+      "required_token_positions_per_second": 27768.137499196208,
+      "required_pages_per_second_upper_estimate": 0.34478733376207465
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14102911641430182,
-    "producer_p95_seconds": 0.3827710149998893,
-    "gpu_update_seconds": 0.9886334285444122,
-    "producer_batches_per_second": 7.090734349226835,
-    "gpu_batches_per_second": 1.0114972558355861,
-    "producer_headroom": 7.010137010573757,
+    "producer_batch_seconds": 0.2289762241794051,
+    "producer_p95_seconds": 0.379161805001786,
+    "gpu_update_seconds": 0.9956735485338299,
+    "producer_batches_per_second": 4.367265656439903,
+    "gpu_batches_per_second": 1.0043452509836592,
+    "producer_headroom": 4.348370893537444,
     "sustainable": true,
-    "buffer_seconds": 60.30663914120914,
+    "buffer_seconds": 60.73608646056363,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 463921.03633565013,
-    "upload_bytes_per_second": 6166340.979345969,
+    "generation_bytes_per_second": 458331.2084130285,
+    "upload_bytes_per_second": 6305182.925547497,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.54665486862305,
+    "estimated_full_shards_per_hour": 98.34720792096273,
     "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": 0.9850222569984908,
-      "bytes": 2356425,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0524133049984812,
       "bytes": 12178249,
@@ -562,6 +556,12 @@
       "bytes": 7094864,
       "pages": 41,
       "shards": 2
+    },
+    {
+      "seconds": 1.0156349210010376,
+      "bytes": 12015881,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -322,3 +322,4 @@
 {"step": 32200, "loss": 0.6732552777975798, "lr": 0.0003, "grad_norm": 0.14508445797455272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4712325524, "tokens_per_second": 150656.4921985203, "data_wait_s": 9.315997886005789e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 32300, "loss": 0.6117614042013884, "lr": 0.0003, "grad_norm": 0.13833622898780742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4727005304, "tokens_per_second": 149149.09986046032, "data_wait_s": 1.377000444335863e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 32400, "loss": 0.6780863795429468, "lr": 0.0003, "grad_norm": 0.12556539759495675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4741489249, "tokens_per_second": 147126.84428959162, "data_wait_s": 9.2410045908764e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 32500, "loss": 0.6430358495563269, "lr": 0.0003, "grad_norm": 0.13856002775264428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4756103473, "tokens_per_second": 149169.28847649784, "data_wait_s": 1.808099477784708e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:389bb6026b59b1be24c2e4a010b75c4ba182b90dc021cea4f3e5390ea2e437c3
-size 685321
+oid sha256:0eacd1d660a73bcdadb651127c3b3945056c3246cd1a63cff8ddc9f7a490751e
+size 687439