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": 75702.79708846263,
-      "required_pages_per_second_upper_estimate": 0.006561726584642853
+      "required_token_positions_per_second": 74851.18918937142,
+      "required_pages_per_second_upper_estimate": 0.0064879113703301145
     },
     {
       "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": 9462.849636057828,
-      "required_pages_per_second_upper_estimate": 0.0037748722020335998
+      "required_token_positions_per_second": 9356.398648671428,
+      "required_pages_per_second_upper_estimate": 0.0037324073115810705
     },
     {
       "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": 18925.699272115657,
-      "required_pages_per_second_upper_estimate": 0.004378127445672894
+      "required_token_positions_per_second": 18712.797297342855,
+      "required_pages_per_second_upper_estimate": 0.004328876320755988
     },
     {
       "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": 28388.54890817349,
-      "required_pages_per_second_upper_estimate": 0.0017287136853025816
+      "required_token_positions_per_second": 28069.195946014283,
+      "required_pages_per_second_upper_estimate": 0.0017092667654225878
     },
     {
       "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": 28388.54890817349,
-      "required_pages_per_second_upper_estimate": 0.002185429079484207
+      "required_token_positions_per_second": 28069.195946014283,
+      "required_pages_per_second_upper_estimate": 0.0021608444044316118
     },
     {
       "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": 28388.54890817349,
-      "required_pages_per_second_upper_estimate": 0.35249076707815646
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28069.195946014283,
+      "required_pages_per_second_upper_estimate": 0.34852547209374923
     }
   ],
   "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": 0.07003387895313296,
-    "producer_p95_seconds": 0.18717053699947428,
-    "gpu_update_seconds": 0.9739138160753165,
-    "producer_batches_per_second": 14.278803558335035,
-    "gpu_batches_per_second": 1.0267848997458582,
-    "producer_headroom": 13.906324062487881,
+    "producer_batch_seconds": 0.05930502062494725,
+    "producer_p95_seconds": 0.21414185399953567,
+    "gpu_update_seconds": 0.9849943708104653,
+    "producer_batches_per_second": 16.86197879137639,
+    "gpu_batches_per_second": 1.0152342283714657,
+    "producer_headroom": 16.6089541902312,
     "sustainable": true,
-    "buffer_seconds": 60.38265659666962,
+    "buffer_seconds": 60.084656619438384,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406831.32041608664,
-    "upload_bytes_per_second": 5942719.262317403,
+    "generation_bytes_per_second": 369186.11055891396,
+    "upload_bytes_per_second": 5937973.585928265,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.29653081285429,
+    "estimated_full_shards_per_hour": 79.21874511313976,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 2564748,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9853730400000131,
+      "bytes": 5552102,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -70,3 +70,4 @@
 {"step": 88900, "loss": 0.6369328573346138, "lr": 0.0003, "grad_norm": 0.13368805654694568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13010034356, "tokens_per_second": 149093.77848599263, "data_wait_s": 9.379000402987003e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 89000, "loss": 0.6425842028111219, "lr": 0.0003, "grad_norm": 0.15072500295912583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13024704107, "tokens_per_second": 149202.7943761418, "data_wait_s": 1.308999981119996e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 89100, "loss": 0.5993454270064831, "lr": 0.0003, "grad_norm": 0.13591362712625177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13039297606, "tokens_per_second": 145374.30608773872, "data_wait_s": 1.313800021307543e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 89200, "loss": 0.6013900879770517, "lr": 0.0003, "grad_norm": 0.18664570087347193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13053837626, "tokens_per_second": 148020.184680706, "data_wait_s": 1.192800027638441e-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 sha256:4a019609ca551c454278958220ef700fc64221e5765fd583bad8aff3480b6dc8
-size 153444
+oid sha256:30da922a0a42c2ebc4b011db2a7b0efcfa6f354980d6bc3a72c949861f6fe449
+size 155562