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": 76114.8255708102,
-      "required_pages_per_second_upper_estimate": 0.006597440169216099
+      "required_token_positions_per_second": 76081.57893027576,
+      "required_pages_per_second_upper_estimate": 0.006594558434677414
     },
     {
       "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": 9514.353196351274,
-      "required_pages_per_second_upper_estimate": 0.0037954177422815038
+      "required_token_positions_per_second": 9510.19736628447,
+      "required_pages_per_second_upper_estimate": 0.003793759919532659
     },
     {
       "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": 19028.70639270255,
-      "required_pages_per_second_upper_estimate": 0.004401956330157266
+      "required_token_positions_per_second": 19020.39473256894,
+      "required_pages_per_second_upper_estimate": 0.0044000335738655786
     },
     {
       "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": 28543.059589053824,
-      "required_pages_per_second_upper_estimate": 0.0017381225751133022
+      "required_token_positions_per_second": 28530.592098853413,
+      "required_pages_per_second_upper_estimate": 0.001737363370371963
     },
     {
       "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": 28543.059589053824,
-      "required_pages_per_second_upper_estimate": 0.0021973237394113145
+      "required_token_positions_per_second": 28530.592098853413,
+      "required_pages_per_second_upper_estimate": 0.002196363957503465
     },
     {
       "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": 28543.059589053824,
-      "required_pages_per_second_upper_estimate": 0.3544092726207063
+      "required_token_positions_per_second": 28530.592098853413,
+      "required_pages_per_second_upper_estimate": 0.35425446811842276
     }
   ],
   "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": 0.07323900700006902,
-    "producer_p95_seconds": 0.19240677400011919,
-    "gpu_update_seconds": 0.9686417783538147,
-    "producer_batches_per_second": 13.653926247239502,
-    "gpu_batches_per_second": 1.0323733937013102,
-    "producer_headroom": 13.225763401637899,
+    "producer_batch_seconds": 0.16005694429686912,
+    "producer_p95_seconds": 0.23078480399999535,
+    "gpu_update_seconds": 0.9690650619589181,
+    "producer_batches_per_second": 6.2477764047852125,
+    "gpu_batches_per_second": 1.031922457279131,
+    "producer_headroom": 6.0545018288086485,
     "sustainable": true,
-    "buffer_seconds": 60.055790257936515,
+    "buffer_seconds": 60.08203384145292,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 172,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 574511.2843118588,
-    "upload_bytes_per_second": 6087022.566980187,
+    "generation_bytes_per_second": 560834.1998339917,
+    "upload_bytes_per_second": 6038393.070529687,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.63837383755123,
+    "estimated_full_shards_per_hour": 60.17098186619193,
     "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": 0.9901489570002013,
-      "bytes": 9709695,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.8841591720001816,
       "bytes": 855761,
@@ -562,6 +556,12 @@
       "bytes": 5837515,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.8290303609992407,
+      "bytes": 5336251,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -129,3 +129,4 @@
 {"step": 52900, "loss": 0.6748346481472254, "lr": 0.0003, "grad_norm": 0.12579535463662597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7742525265, "tokens_per_second": 150655.0856055337, "data_wait_s": 1.0423998901387677e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 53000, "loss": 0.7239194959402084, "lr": 0.0003, "grad_norm": 0.1563958009252826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7757197901, "tokens_per_second": 150552.44047842812, "data_wait_s": 1.190599869005382e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 53100, "loss": 0.6875843312591314, "lr": 0.0003, "grad_norm": 0.14041930114467732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7771892107, "tokens_per_second": 150686.94031380702, "data_wait_s": 1.6582998796366155e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 53200, "loss": 0.6077895984053612, "lr": 0.0003, "grad_norm": 0.1264437712783856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7786437542, "tokens_per_second": 148231.86932768498, "data_wait_s": 1.2634000086109154e-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 sha256:5fd34bfc7aefe0910d882ad0296700c4202391d8d145702e67d310fdeef4d32e
-size 279189
+oid sha256:0ee4873e52de5f417a97da689f6b2ee807c9c34871dd01a91d14be436e9590a4
+size 281307