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": 76739.50495878958,
-      "required_pages_per_second_upper_estimate": 0.005309017013232473
+      "required_token_positions_per_second": 76244.18377252348,
+      "required_pages_per_second_upper_estimate": 0.005274749544263089
     },
     {
       "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": 9592.438119848697,
-      "required_pages_per_second_upper_estimate": 0.0040161535278943694
+      "required_token_positions_per_second": 9530.522971565435,
+      "required_pages_per_second_upper_estimate": 0.003990230948243488
     },
     {
       "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": 19184.876239697394,
-      "required_pages_per_second_upper_estimate": 0.003995703781519588
+      "required_token_positions_per_second": 19061.04594313087,
+      "required_pages_per_second_upper_estimate": 0.003969913196369306
     },
     {
       "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": 28777.314359546093,
-      "required_pages_per_second_upper_estimate": 0.0017251560554399938
+      "required_token_positions_per_second": 28591.5689146963,
+      "required_pages_per_second_upper_estimate": 0.001714020899638121
     },
     {
       "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": 28777.314359546093,
-      "required_pages_per_second_upper_estimate": 0.0022373096134260194
+      "required_token_positions_per_second": 28591.5689146963,
+      "required_pages_per_second_upper_estimate": 0.002222868722096816
     },
     {
       "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": 28777.314359546093,
-      "required_pages_per_second_upper_estimate": 0.3691244899314541
+      "required_token_positions_per_second": 28591.5689146963,
+      "required_pages_per_second_upper_estimate": 0.36674194680284117
     }
   ],
   "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": 0.24709898691381227,
-    "producer_p95_seconds": 0.2840462260028289,
-    "gpu_update_seconds": 0.9607567841308488,
-    "producer_batches_per_second": 4.046961149010289,
-    "gpu_batches_per_second": 1.0408461501571937,
-    "producer_headroom": 3.8881453790256097,
+    "producer_batch_seconds": 0.26639515541393166,
+    "producer_p95_seconds": 0.2920580220015836,
+    "gpu_update_seconds": 0.9669983512443313,
+    "producer_batches_per_second": 3.7538220184453963,
+    "gpu_batches_per_second": 1.0341279266021521,
+    "producer_headroom": 3.6299397027013662,
     "sustainable": true,
-    "buffer_seconds": 60.52767740024347,
+    "buffer_seconds": 60.920896128392876,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 162,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 561778.0533472627,
-    "upload_bytes_per_second": 5092282.7306850925,
+    "generation_bytes_per_second": 536704.605522428,
+    "upload_bytes_per_second": 5205483.424044281,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.272246362273265,
+    "estimated_full_shards_per_hour": 115.16431450132973,
     "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": 1.1945516140003747,
-      "bytes": 6236357,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1350728899997193,
       "bytes": 701024,
@@ -562,6 +556,12 @@
       "bytes": 6126927,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3508422129998507,
+      "bytes": 16469825,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -327,3 +327,4 @@
 {"step": 752700, "loss": 0.5622467994689941, "lr": 5e-05, "grad_norm": 0.27329237892658775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110174321758, "tokens_per_second": 152240.46260124017, "data_wait_s": 1.2046999472659081e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 752800, "loss": 0.5367382168769836, "lr": 5e-05, "grad_norm": 0.24947319585303135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110188812076, "tokens_per_second": 149083.8764930113, "data_wait_s": 1.9612001779023558e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 752900, "loss": 0.5307222604751587, "lr": 5e-05, "grad_norm": 0.2496026372006396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110203286186, "tokens_per_second": 149422.99089987634, "data_wait_s": 1.1863998224725947e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 753000, "loss": 0.5361687541007996, "lr": 5e-05, "grad_norm": 0.24043082826037804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110217965101, "tokens_per_second": 151532.8745609649, "data_wait_s": 1.5124998753890395e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:805e7de38a7a1592d3e87e203f6957271e0aa439b2cd180414e690eba742974b
-size 701685
+oid sha256:9cb2a9bd8d376c3005c44de90872b1f9197b747730f2f3081ede27ba267dde2c
+size 703803