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": 74661.84648542074,
-      "required_pages_per_second_upper_estimate": 0.0064714996246363655
+      "required_token_positions_per_second": 75120.75553690466,
+      "required_pages_per_second_upper_estimate": 0.006511276698124634
     },
     {
       "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": 9332.730810677593,
-      "required_pages_per_second_upper_estimate": 0.0037229658571396175
+      "required_token_positions_per_second": 9390.094442113083,
+      "required_pages_per_second_upper_estimate": 0.003745849067381954
     },
     {
       "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": 18665.461621355185,
-      "required_pages_per_second_upper_estimate": 0.0043179260451956854
+      "required_token_positions_per_second": 18780.188884226165,
+      "required_pages_per_second_upper_estimate": 0.004344466178330022
     },
     {
       "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": 27998.192432032778,
-      "required_pages_per_second_upper_estimate": 0.0017049430239477617
+      "required_token_positions_per_second": 28170.28332633925,
+      "required_pages_per_second_upper_estimate": 0.0017154224565198825
     },
     {
       "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": 27998.192432032778,
-      "required_pages_per_second_upper_estimate": 0.0021553783573750093
+      "required_token_positions_per_second": 28170.28332633925,
+      "required_pages_per_second_upper_estimate": 0.0021686263907968063
     },
     {
       "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": 27998.192432032778,
-      "required_pages_per_second_upper_estimate": 0.34764384608357374
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28170.28332633925,
+      "required_pages_per_second_upper_estimate": 0.3497806390396867
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3960835619999443,
-    "producer_p95_seconds": 0.29465220400015824,
-    "gpu_update_seconds": 0.9874923199816241,
-    "producer_batches_per_second": 2.524719771127842,
-    "gpu_batches_per_second": 1.0126661035891487,
-    "producer_headroom": 2.4931413840945083,
+    "producer_batch_seconds": 0.25856511172651153,
+    "producer_p95_seconds": 0.24305080699923565,
+    "gpu_update_seconds": 0.9814597773018877,
+    "producer_batches_per_second": 3.867497797064424,
+    "gpu_batches_per_second": 1.0188904559584508,
+    "producer_headroom": 3.795793526622391,
     "sustainable": true,
-    "buffer_seconds": 60.23703151887907,
+    "buffer_seconds": 60.85050619271704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369410.50482843834,
-    "upload_bytes_per_second": 5912253.735729597,
+    "generation_bytes_per_second": 358167.0562896291,
+    "upload_bytes_per_second": 6007747.908546246,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.26689489974845,
+    "estimated_full_shards_per_hour": 76.85431257740645,
     "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": 1.2463584760007507,
-      "bytes": 4721254,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.214833362000718,
       "bytes": 7500577,
@@ -562,6 +556,12 @@
       "bytes": 5699054,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 1.2686124620013288,
+      "bytes": 11642195,
+      "pages": 44,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -146,3 +146,4 @@
 {"step": 99600, "loss": 0.6056494414806366, "lr": 0.0003, "grad_norm": 0.1452663709025324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14574631938, "tokens_per_second": 147439.03541988644, "data_wait_s": 1.087899727281183e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 99700, "loss": 0.5871254727244377, "lr": 0.0003, "grad_norm": 0.14458459120061495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14589058669, "tokens_per_second": 137545.56160873885, "data_wait_s": 1.1546999303391203e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 99800, "loss": 0.643492616713047, "lr": 0.0003, "grad_norm": 0.16165238422953543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14603748246, "tokens_per_second": 149025.5432800379, "data_wait_s": 1.6723999578971416e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 99900, "loss": 0.6030363850295544, "lr": 0.0003, "grad_norm": 0.12983061304929588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14618405878, "tokens_per_second": 148787.18523982857, "data_wait_s": 1.7555998056195676e-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 sha256:876c2ace1ded5a0d482ca8ee95f2dd6dca10125abbe584cf2ceade0f0a3646a7
-size 315195
+oid sha256:b1be8b9e186e9e8a73be6ed3d31e1915285b7ce873d9a2c0c1a6d89a08def9fe
+size 317313