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": 75686.57593504993,
-      "required_pages_per_second_upper_estimate": 0.006560320576177178
+      "required_token_positions_per_second": 74280.85185318752,
+      "required_pages_per_second_upper_estimate": 0.006438475975536457
     },
     {
       "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": 9460.821991881241,
-      "required_pages_per_second_upper_estimate": 0.003774063344455577
+      "required_token_positions_per_second": 9285.10648164844,
+      "required_pages_per_second_upper_estimate": 0.0037039678002427157
     },
     {
       "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": 18921.643983762482,
-      "required_pages_per_second_upper_estimate": 0.004377189326083024
+      "required_token_positions_per_second": 18570.21296329688,
+      "required_pages_per_second_upper_estimate": 0.0042958919444730246
     },
     {
       "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": 28382.465975643725,
-      "required_pages_per_second_upper_estimate": 0.0017283432666261997
+      "required_token_positions_per_second": 27855.31944494532,
+      "required_pages_per_second_upper_estimate": 0.0016962428086307667
     },
     {
       "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": 28382.465975643725,
-      "required_pages_per_second_upper_estimate": 0.0021849607984994308
+      "required_token_positions_per_second": 27855.31944494532,
+      "required_pages_per_second_upper_estimate": 0.0021443795993312752
     },
     {
       "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": 28382.465975643725,
-      "required_pages_per_second_upper_estimate": 0.3524152374144024
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27855.31944494532,
+      "required_pages_per_second_upper_estimate": 0.3458698417490758
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.17048431457030944,
-    "producer_p95_seconds": 0.14274176300000363,
-    "gpu_update_seconds": 0.9741225453674813,
-    "producer_batches_per_second": 5.86564225876387,
-    "gpu_batches_per_second": 1.0265648862718362,
-    "producer_headroom": 5.713854367322123,
+    "producer_batch_seconds": 0.04359676914063115,
+    "producer_p95_seconds": 0.1390367519998108,
+    "gpu_update_seconds": 0.9925572763451851,
+    "producer_batches_per_second": 22.937479536024235,
+    "gpu_batches_per_second": 1.0074985331649784,
+    "producer_headroom": 22.766762214499636,
     "sustainable": true,
-    "buffer_seconds": 60.39559781278384,
+    "buffer_seconds": 60.54599385705629,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 73,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 489976.10101581464,
-    "upload_bytes_per_second": 4557104.891474862,
+    "generation_bytes_per_second": 486829.262728603,
+    "upload_bytes_per_second": 4919206.073839564,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.13746521812276,
+    "estimated_full_shards_per_hour": 104.4622269763333,
     "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": [
@@ -250,6 +250,12 @@
       "bytes": 2623386,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1859510890001275,
+      "bytes": 10202108,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -11,3 +11,4 @@
 {"step": 83000, "loss": 0.6183635629713535, "lr": 0.0003, "grad_norm": 0.13188186813879005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12146491385, "tokens_per_second": 150376.761259858, "data_wait_s": 1.3017000128456857e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 83100, "loss": 0.6229391451925039, "lr": 0.0003, "grad_norm": 0.13720698418826618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12160969657, "tokens_per_second": 146814.05009065697, "data_wait_s": 1.158499981102068e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 83200, "loss": 0.6312891934067011, "lr": 0.0003, "grad_norm": 0.13139326862963113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12175734091, "tokens_per_second": 149885.2611167311, "data_wait_s": 8.77899992701714e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 83300, "loss": 0.7064251564443111, "lr": 0.0003, "grad_norm": 0.16103747510548544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12190469472, "tokens_per_second": 149129.6152356179, "data_wait_s": 1.8023000166067504e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3457b4c118de80cd15b8a87fe96206392c4e866112c7cd4bd92658f8792ab263
-size 27699
+oid sha256:d98d844b0c2503a4ba208ce345ab69b419f5251eb850bee1db61e3e029ed0730
+size 29817