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": 74856.21750028821,
-      "required_pages_per_second_upper_estimate": 0.006488347211576245
+      "required_token_positions_per_second": 75666.74173363415,
+      "required_pages_per_second_upper_estimate": 0.006558601397867782
     },
     {
       "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": 9357.027187536027,
-      "required_pages_per_second_upper_estimate": 0.0037326580451316526
+      "required_token_positions_per_second": 9458.342716704268,
+      "required_pages_per_second_upper_estimate": 0.003773074324519016
     },
     {
       "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": 18714.054375072054,
-      "required_pages_per_second_upper_estimate": 0.004329167123564826
+      "required_token_positions_per_second": 18916.685433408536,
+      "required_pages_per_second_upper_estimate": 0.004376042252726145
     },
     {
       "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": 28071.08156260808,
-      "required_pages_per_second_upper_estimate": 0.0017093815895800845
+      "required_token_positions_per_second": 28375.028150112805,
+      "required_pages_per_second_upper_estimate": 0.0017278903420746212
     },
     {
       "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": 28071.08156260808,
-      "required_pages_per_second_upper_estimate": 0.0021609895644167233
+      "required_token_positions_per_second": 28375.028150112805,
+      "required_pages_per_second_upper_estimate": 0.0021843882141008414
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28071.08156260808,
-      "required_pages_per_second_upper_estimate": 0.34854888514109145
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28375.028150112805,
+      "required_pages_per_second_upper_estimate": 0.3523228845141091
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07402864277304388,
-    "producer_p95_seconds": 0.26244286200017086,
-    "gpu_update_seconds": 0.9849282058596687,
-    "producer_batches_per_second": 13.508284935950911,
-    "gpu_batches_per_second": 1.0153024292031279,
-    "producer_headroom": 13.30469084620732,
+    "producer_batch_seconds": 0.20268086643741867,
+    "producer_p95_seconds": 0.30183024700090755,
+    "gpu_update_seconds": 0.9743778879701336,
+    "producer_batches_per_second": 4.933864836761825,
+    "gpu_batches_per_second": 1.0262958676979457,
+    "producer_headroom": 4.807448799174095,
     "sustainable": true,
-    "buffer_seconds": 60.08062055743979,
+    "buffer_seconds": 60.41142905414828,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 171,
+    "upload_pages_per_commit": 168,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 624476.1912468428,
-    "upload_bytes_per_second": 5161133.852900171,
+    "generation_bytes_per_second": 602030.9923572884,
+    "upload_bytes_per_second": 5084854.176124425,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.99902678992254,
+    "estimated_full_shards_per_hour": 64.59091819781774,
     "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": 1.112150704000669,
-      "bytes": 16576455,
-      "pages": 44,
-      "shards": 4
-    },
     {
       "seconds": 0.7744886969994695,
       "bytes": 1014054,
@@ -562,6 +556,12 @@
       "bytes": 7335325,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.085818534000282,
+      "bytes": 10112362,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -233,3 +233,4 @@
 {"step": 63300, "loss": 0.6152825839817524, "lr": 0.0003, "grad_norm": 0.13509051985736967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9264031394, "tokens_per_second": 149876.53349747547, "data_wait_s": 1.4135999663267285e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 63400, "loss": 0.7246140874922276, "lr": 0.0003, "grad_norm": 0.17346604542955618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9278631187, "tokens_per_second": 148911.84665287365, "data_wait_s": 1.4029999874765053e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 63500, "loss": 0.6298396866768599, "lr": 0.0003, "grad_norm": 0.13247659515158774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9293284842, "tokens_per_second": 149255.57523236945, "data_wait_s": 1.6213998605962843e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 63600, "loss": 0.6364936083555222, "lr": 0.0003, "grad_norm": 0.13461042153087086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9307954687, "tokens_per_second": 149933.33936021413, "data_wait_s": 1.1206000635866076e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:44a703a5d3537c212d456810e6d6c9d57bd44053133414248625622f33ca97ad
-size 501027
+oid sha256:b8bd8b6e7b01d9c693cb98420f6146791f8139cde048bc253a3bc6c97ccae1d0
+size 503145