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": 76721.90983312913,
-      "required_pages_per_second_upper_estimate": 0.005307799741612967
+      "required_token_positions_per_second": 76841.83261226915,
+      "required_pages_per_second_upper_estimate": 0.005316096277733058
     },
     {
       "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": 9590.238729141141,
-      "required_pages_per_second_upper_estimate": 0.004015232688933616
+      "required_token_positions_per_second": 9605.229076533644,
+      "required_pages_per_second_upper_estimate": 0.004021508834352807
     },
     {
       "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": 19180.477458282283,
-      "required_pages_per_second_upper_estimate": 0.003994787631354487
+      "required_token_positions_per_second": 19210.458153067288,
+      "required_pages_per_second_upper_estimate": 0.004001031819434114
     },
     {
       "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": 28770.716187423426,
-      "required_pages_per_second_upper_estimate": 0.0017247605050960653
+      "required_token_positions_per_second": 28815.68722960093,
+      "required_pages_per_second_upper_estimate": 0.0017274564504078014
     },
     {
       "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": 28770.716187423426,
-      "required_pages_per_second_upper_estimate": 0.002236796634565774
+      "required_token_positions_per_second": 28815.68722960093,
+      "required_pages_per_second_upper_estimate": 0.0022402929352883655
     },
     {
       "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": 28770.716187423426,
-      "required_pages_per_second_upper_estimate": 0.3690398556640298
+      "required_token_positions_per_second": 28815.68722960093,
+      "required_pages_per_second_upper_estimate": 0.3696166959069398
     }
   ],
   "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": 0.3170916938672157,
-    "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,
+    "producer_batch_seconds": 0.0636467814296715,
+    "producer_p95_seconds": 0.1424965650003287,
+    "gpu_update_seconds": 0.9594773770170082,
+    "producer_batches_per_second": 15.711713578242462,
+    "gpu_batches_per_second": 1.0422340577835985,
+    "producer_headroom": 15.07503373249459,
     "sustainable": true,
-    "buffer_seconds": 60.54155859913579,
+    "buffer_seconds": 60.447074752071515,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 414773.35473780835,
-    "upload_bytes_per_second": 5215596.187207662,
+    "generation_bytes_per_second": 398238.6562341637,
+    "upload_bytes_per_second": 5373177.55584162,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.00070649719895,
+    "estimated_full_shards_per_hour": 85.45274510639841,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 7369290,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1581361090002247,
+      "bytes": 10067921,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -24,3 +24,4 @@
 {"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}
+{"step": 559800, "loss": 0.6557629704475403, "lr": 0.0003, "grad_norm": 0.20263330391709677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81936917792, "tokens_per_second": 151485.46904864797, "data_wait_s": 1.386499980071676e-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 sha256:92a36fb2348cacfb85ac828581cc59461cd399be57a93d41d561ec5cbf654ebe
-size 55233
+oid sha256:29d62befc52afe840bee15fa793a1f7bd17e7d68e569a64e46e836e94c8ee7c2
+size 57351