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": 74714.78188742291,
-      "required_pages_per_second_upper_estimate": 0.006476087931118364
+      "required_token_positions_per_second": 74924.06948841283,
+      "required_pages_per_second_upper_estimate": 0.006494228449937595
     },
     {
       "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": 9339.347735927864,
-      "required_pages_per_second_upper_estimate": 0.0037256054475537995
+      "required_token_positions_per_second": 9365.508686051604,
+      "required_pages_per_second_upper_estimate": 0.0037360414416992196
     },
     {
       "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": 18678.695471855728,
-      "required_pages_per_second_upper_estimate": 0.004320987463601171
+      "required_token_positions_per_second": 18731.017372103208,
+      "required_pages_per_second_upper_estimate": 0.004333091214389432
     },
     {
       "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": 28018.04320778359,
-      "required_pages_per_second_upper_estimate": 0.0017061518320420149
+      "required_token_positions_per_second": 28096.52605815481,
+      "required_pages_per_second_upper_estimate": 0.0017109310258619293
     },
     {
       "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": 28018.04320778359,
-      "required_pages_per_second_upper_estimate": 0.002156906525042771
+      "required_token_positions_per_second": 28096.52605815481,
+      "required_pages_per_second_upper_estimate": 0.002162948352118782
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28018.04320778359,
-      "required_pages_per_second_upper_estimate": 0.3478903262821261
+      "required_token_positions_per_second": 28096.52605815481,
+      "required_pages_per_second_upper_estimate": 0.34886482061853324
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08302199285152767,
-    "producer_p95_seconds": 0.18523669699970924,
-    "gpu_update_seconds": 0.9867926819500089,
-    "producer_batches_per_second": 12.045001157564952,
-    "gpu_batches_per_second": 1.0133840859296728,
-    "producer_headroom": 11.885918996364481,
+    "producer_batch_seconds": 0.07519265721882107,
+    "producer_p95_seconds": 0.18400543099960487,
+    "gpu_update_seconds": 0.9840362450067157,
+    "producer_batches_per_second": 13.29917091624866,
+    "gpu_batches_per_second": 1.0162227306913632,
+    "producer_headroom": 13.086866210127853,
     "sustainable": true,
-    "buffer_seconds": 60.19435359895054,
+    "buffer_seconds": 60.02621094540966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312914.2843067584,
+    "generation_bytes_per_second": 345064.2240766591,
     "upload_bytes_per_second": 5776035.999859456,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.14412114049972,
+    "estimated_full_shards_per_hour": 74.04274980282622,
     "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": [

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -29,3 +29,4 @@
 {"step": 283100, "loss": 0.5726057942956686, "lr": 0.0003, "grad_norm": 0.14841560961660563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41437897356, "tokens_per_second": 128897.20193413284, "data_wait_s": 1.4055998690309934e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 283200, "loss": 0.60996538028121, "lr": 0.0003, "grad_norm": 0.16046535164338369, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41452543101, "tokens_per_second": 150093.66637814394, "data_wait_s": 9.75399962044321e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 283300, "loss": 0.5941618494689465, "lr": 0.0003, "grad_norm": 0.14795023479137123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41467242427, "tokens_per_second": 150159.80292535602, "data_wait_s": 1.1094000001321547e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 283400, "loss": 0.5975897945463657, "lr": 0.0003, "grad_norm": 0.16067724468399516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41481867567, "tokens_per_second": 150081.94934486123, "data_wait_s": 1.3803999536321498e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6690dc6abf6b5b88607b24377e206b1511558e944e81686768e393c626a6b15e
-size 65823
+oid sha256:e41132b59e584c3cc299cb818569129a1414803ac38bfdeedc49e648fc3566c4
+size 67941