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": 75806.09564950923,
-      "required_pages_per_second_upper_estimate": 0.006570680242106563
+      "required_token_positions_per_second": 74949.47492182034,
+      "required_pages_per_second_upper_estimate": 0.006496430528515875
     },
     {
       "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": 9475.761956188653,
-      "required_pages_per_second_upper_estimate": 0.003780023119590176
+      "required_token_positions_per_second": 9368.684365227542,
+      "required_pages_per_second_upper_estimate": 0.003737308267603136
     },
     {
       "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": 18951.523912377306,
-      "required_pages_per_second_upper_estimate": 0.0043841015216992815
+      "required_token_positions_per_second": 18737.368730455084,
+      "required_pages_per_second_upper_estimate": 0.00433456048936405
     },
     {
       "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": 28427.285868565956,
-      "required_pages_per_second_upper_estimate": 0.0017310725629533588
+      "required_token_positions_per_second": 28106.05309568263,
+      "required_pages_per_second_upper_estimate": 0.0017115111724628706
     },
     {
       "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": 28427.285868565956,
-      "required_pages_per_second_upper_estimate": 0.0021884111579259886
+      "required_token_positions_per_second": 28106.05309568263,
+      "required_pages_per_second_upper_estimate": 0.0021636817698402013
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28427.285868565956,
-      "required_pages_per_second_upper_estimate": 0.3529717504819643
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28106.05309568263,
+      "required_pages_per_second_upper_estimate": 0.3489831145396852
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.34209865507023096,
-    "producer_p95_seconds": 0.34485396499803755,
-    "gpu_update_seconds": 0.9725866946225891,
-    "producer_batches_per_second": 2.9231333861710316,
-    "gpu_batches_per_second": 1.0281859761489422,
-    "producer_headroom": 2.84300063799702,
+    "producer_batch_seconds": 0.1060570514379151,
+    "producer_p95_seconds": 0.3466564699992887,
+    "gpu_update_seconds": 0.9837026887367195,
+    "producer_batches_per_second": 9.428887437865379,
+    "gpu_batches_per_second": 1.0165673139352802,
+    "producer_headroom": 9.27522192442405,
     "sustainable": true,
-    "buffer_seconds": 60.30037506660052,
+    "buffer_seconds": 60.00586401293989,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310836.34175566764,
-    "upload_bytes_per_second": 6123685.411910331,
+    "generation_bytes_per_second": 292643.323498026,
+    "upload_bytes_per_second": 6030474.122966458,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.69824304106257,
+    "estimated_full_shards_per_hour": 62.79444483476243,
     "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": 1.128464265002549,
-      "bytes": 12438546,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.0890558419996523,
       "bytes": 11718926,
@@ -562,6 +556,12 @@
       "bytes": 11923116,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.3382599670003401,
+      "bytes": 7199956,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -268,3 +268,4 @@
 {"step": 111800, "loss": 0.6815985608845949, "lr": 0.0003, "grad_norm": 0.21084688447450772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16359614931, "tokens_per_second": 148037.53453362366, "data_wait_s": 1.5261000953614712e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 111900, "loss": 0.5819695480167866, "lr": 0.0003, "grad_norm": 0.14860150425731405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16374251401, "tokens_per_second": 149879.69115942155, "data_wait_s": 9.020004654303193e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 112000, "loss": 0.5638746749609709, "lr": 0.0003, "grad_norm": 0.1325248474738872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16388938660, "tokens_per_second": 148220.7544071908, "data_wait_s": 1.442500069970265e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 112100, "loss": 0.5938390232622623, "lr": 0.0003, "grad_norm": 0.1428568940247015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16403572042, "tokens_per_second": 149137.73920208952, "data_wait_s": 1.1523996363393962e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3dfb1d0bf3f659ecb05f80f72964494c36cfa7922b761e9b76e5b9909d95df49
-size 575940
+oid sha256:53617ccea2012bc432837119883b3792265ff7f3d3cae1c174d83a0c5c66dcc5
+size 578058