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": 74882.83766688577,
-      "required_pages_per_second_upper_estimate": 0.006490654580148725
+      "required_token_positions_per_second": 75731.21103212437,
+      "required_pages_per_second_upper_estimate": 0.0065641894332649706
     },
     {
       "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": 9360.35470836072,
-      "required_pages_per_second_upper_estimate": 0.003733985442939493
+      "required_token_positions_per_second": 9466.401379015546,
+      "required_pages_per_second_upper_estimate": 0.0037762890454027228
     },
     {
       "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": 18720.70941672144,
-      "required_pages_per_second_upper_estimate": 0.00433070665032568
+      "required_token_positions_per_second": 18932.80275803109,
+      "required_pages_per_second_upper_estimate": 0.004379770712122348
     },
     {
       "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": 28081.064125082165,
-      "required_pages_per_second_upper_estimate": 0.0017099894752602986
+      "required_token_positions_per_second": 28399.20413704664,
+      "required_pages_per_second_upper_estimate": 0.0017293625328372911
     },
     {
       "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": 28081.064125082165,
-      "required_pages_per_second_upper_estimate": 0.0021617580497094798
+      "required_token_positions_per_second": 28399.20413704664,
+      "required_pages_per_second_upper_estimate": 0.002186249348498423
     },
     {
       "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": 28081.064125082165,
-      "required_pages_per_second_upper_estimate": 0.3486728351575321
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28399.20413704664,
+      "required_pages_per_second_upper_estimate": 0.352623069360004
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.10149520252340949,
-    "producer_p95_seconds": 0.2266594669999904,
-    "gpu_update_seconds": 0.9845780728553178,
-    "producer_batches_per_second": 9.852682443481541,
-    "gpu_batches_per_second": 1.0156634883203908,
-    "producer_headroom": 9.70073509265848,
+    "producer_batch_seconds": 0.19920030615637074,
+    "producer_p95_seconds": 0.24541962999865063,
+    "gpu_update_seconds": 0.9735484088419684,
+    "producer_batches_per_second": 5.0200726057870995,
+    "gpu_batches_per_second": 1.027170288521652,
+    "producer_headroom": 4.887283697635184,
     "sustainable": true,
-    "buffer_seconds": 60.05926244417439,
+    "buffer_seconds": 60.36000134820204,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420563.12230035535,
-    "upload_bytes_per_second": 5959846.25985183,
+    "generation_bytes_per_second": 442214.1127743015,
+    "upload_bytes_per_second": 5984337.118994301,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.24305583723064,
+    "estimated_full_shards_per_hour": 94.88885438367637,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 5916543,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0784175260014308,
+      "bytes": 7549095,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -97,3 +97,4 @@
 {"step": 94700, "loss": 0.6323264893144369, "lr": 0.0003, "grad_norm": 0.13136582361796362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13858704377, "tokens_per_second": 149837.64433980503, "data_wait_s": 1.2387001333991066e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 94800, "loss": 0.6088749971240759, "lr": 0.0003, "grad_norm": 0.13870514125347966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13873367436, "tokens_per_second": 149842.19417131858, "data_wait_s": 9.170002158498392e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 94900, "loss": 0.6596100553870201, "lr": 0.0003, "grad_norm": 0.14721265667715122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13888128543, "tokens_per_second": 151242.02691629334, "data_wait_s": 1.3546003174269572e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 95000, "loss": 0.6396308094263077, "lr": 0.0003, "grad_norm": 0.143320235435905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13902633353, "tokens_per_second": 148697.1748974744, "data_wait_s": 1.7082998965634033e-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 sha256:2b66ad1c8635b1ec5e341e81a933a87e92cbb4aa7fa42b04a471437dc542c05d
-size 210630
+oid sha256:6df405ea109319c17c00ea78d85f073ec1a8149a208071ce09433e7db1b9f3a7
+size 212748