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": 74726.12750727916,
-      "required_pages_per_second_upper_estimate": 0.006477071340692293
+      "required_token_positions_per_second": 75116.88919282005,
+      "required_pages_per_second_upper_estimate": 0.00651094157321854
     },
     {
       "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": 9340.765938409895,
-      "required_pages_per_second_upper_estimate": 0.0037261711897278982
+      "required_token_positions_per_second": 9389.611149102506,
+      "required_pages_per_second_upper_estimate": 0.0037456562745741607
     },
     {
       "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": 18681.53187681979,
-      "required_pages_per_second_upper_estimate": 0.004321643615970585
+      "required_token_positions_per_second": 18779.22229820501,
+      "required_pages_per_second_upper_estimate": 0.004344242575665361
     },
     {
       "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": 28022.297815229686,
-      "required_pages_per_second_upper_estimate": 0.001706410915313282
+      "required_token_positions_per_second": 28168.83344730752,
+      "required_pages_per_second_upper_estimate": 0.0017153341664937775
     },
     {
       "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": 28022.297815229686,
-      "required_pages_per_second_upper_estimate": 0.0021572340565014785
+      "required_token_positions_per_second": 28168.83344730752,
+      "required_pages_per_second_upper_estimate": 0.0021685147750954217
     },
     {
       "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": 28022.297815229686,
-      "required_pages_per_second_upper_estimate": 0.34794315426735145
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28168.83344730752,
+      "required_pages_per_second_upper_estimate": 0.3497626363945456
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.13716007199235491,
-    "producer_p95_seconds": 0.4414159650041256,
-    "gpu_update_seconds": 0.9866428578520687,
-    "producer_batches_per_second": 7.290751495491621,
-    "gpu_batches_per_second": 1.0135379707476015,
-    "producer_headroom": 7.193367891401097,
+    "producer_batch_seconds": 0.1146260939297008,
+    "producer_p95_seconds": 0.4590636499997345,
+    "gpu_update_seconds": 0.9815102940531142,
+    "producer_batches_per_second": 8.724017069039196,
+    "gpu_batches_per_second": 1.0188380153106018,
+    "producer_headroom": 8.562712558757049,
     "sustainable": true,
-    "buffer_seconds": 60.185214328976194,
+    "buffer_seconds": 60.85363823129308,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280441.08105400676,
-    "upload_bytes_per_second": 5431960.895851179,
+    "generation_bytes_per_second": 267728.05276524095,
+    "upload_bytes_per_second": 5410909.444487128,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.17612765994217,
+    "estimated_full_shards_per_hour": 57.44820773332521,
     "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": 1.0266037770052208,
-      "bytes": 7314912,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1214634670031955,
       "bytes": 5616787,
@@ -562,6 +556,12 @@
       "bytes": 1063474,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.102702958000009,
+      "bytes": 6111748,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -420,3 +420,4 @@
 {"step": 357000, "loss": 0.604896292090416, "lr": 0.0003, "grad_norm": 0.15028696367232258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52254457624, "tokens_per_second": 145493.48765809817, "data_wait_s": 1.3530996511690319e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357100, "loss": 0.5408975128084421, "lr": 0.0003, "grad_norm": 0.15872110976417061, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52269172951, "tokens_per_second": 148331.15850647166, "data_wait_s": 1.1898999218828976e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357200, "loss": 0.6047567324712873, "lr": 0.0003, "grad_norm": 0.15845736428146975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52283949229, "tokens_per_second": 149524.7630702091, "data_wait_s": 1.6414000128861517e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 357300, "loss": 0.580042626708746, "lr": 0.0003, "grad_norm": 0.156638364723009, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52298628386, "tokens_per_second": 147576.94735376287, "data_wait_s": 1.0554002074059099e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe6906dde62fb90e08768b129c0ba639e0037edcf4514126ca792143841b0934
-size 900225
+oid sha256:b4f175e91cbca6d7a2c9e2b1e9d401871139c69961674d1807e1b8d28735e2a2
+size 902343