Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76365.49374086368,
-      "required_pages_per_second_upper_estimate": 0.005283142049350771
+      "required_token_positions_per_second": 76721.90983312913,
+      "required_pages_per_second_upper_estimate": 0.005307799741612967
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9545.68671760796,
-      "required_pages_per_second_upper_estimate": 0.003996579692056468
+      "required_token_positions_per_second": 9590.238729141141,
+      "required_pages_per_second_upper_estimate": 0.004015232688933616
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19091.37343521592,
-      "required_pages_per_second_upper_estimate": 0.003976229613180871
+      "required_token_positions_per_second": 19180.477458282283,
+      "required_pages_per_second_upper_estimate": 0.003994787631354487
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28637.060152823877,
-      "required_pages_per_second_upper_estimate": 0.0017167480299027705
+      "required_token_positions_per_second": 28770.716187423426,
+      "required_pages_per_second_upper_estimate": 0.0017247605050960653
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28637.060152823877,
-      "required_pages_per_second_upper_estimate": 0.0022264054657664253
+      "required_token_positions_per_second": 28770.716187423426,
+      "required_pages_per_second_upper_estimate": 0.002236796634565774
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28637.060152823877,
-      "required_pages_per_second_upper_estimate": 0.36732545956085577
+      "required_token_positions_per_second": 28770.716187423426,
+      "required_pages_per_second_upper_estimate": 0.3690398556640298
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09060376662502279,
-    "producer_p95_seconds": 0.1468131239998911,
-    "gpu_update_seconds": 0.9654622315438218,
-    "producer_batches_per_second": 11.037068736211037,
-    "gpu_batches_per_second": 1.035773298351558,
-    "producer_headroom": 10.655873011764857,
+    "producer_batch_seconds": 0.3170916938672157,
+    "producer_p95_seconds": 0.15353993099961372,
+    "gpu_update_seconds": 0.960977120621203,
+    "producer_batches_per_second": 3.153661919693036,
+    "gpu_batches_per_second": 1.0406075009918774,
+    "producer_headroom": 3.0305969509993496,
     "sustainable": true,
-    "buffer_seconds": 60.82412058726077,
+    "buffer_seconds": 60.54155859913579,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432362.713651246,
-    "upload_bytes_per_second": 5142552.309660523,
+    "generation_bytes_per_second": 414773.35473780835,
+    "upload_bytes_per_second": 5215596.187207662,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.77497346070322,
+    "estimated_full_shards_per_hour": 89.00070649719895,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 1125687,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.0864234489999944,
+      "bytes": 7369290,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -23,3 +23,4 @@
 {"step": 559400, "loss": 0.5463201999664307, "lr": 0.0003, "grad_norm": 0.16494876638535577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81878508084, "tokens_per_second": 151462.2509951918, "data_wait_s": 1.4911000107531436e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 559500, "loss": 0.5036600232124329, "lr": 0.0003, "grad_norm": 0.1517686746927503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81893113832, "tokens_per_second": 151727.29383739916, "data_wait_s": 1.885900019260589e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 559600, "loss": 0.5688484907150269, "lr": 0.0003, "grad_norm": 0.17586884351772264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81907705924, "tokens_per_second": 151897.5343654506, "data_wait_s": 1.491900002292823e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 559700, "loss": 0.5579128265380859, "lr": 0.0003, "grad_norm": 0.15052356989310225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81922378235, "tokens_per_second": 152137.26047137222, "data_wait_s": 1.77590000021155e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb5383fa99933433f8ffb3708054823839ef33d7e857f673a807dfb908979ede
-size 53115
+oid sha256:92a36fb2348cacfb85ac828581cc59461cd399be57a93d41d561ec5cbf654ebe
+size 55233