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": 75289.23599358511,
-      "required_pages_per_second_upper_estimate": 0.006525880157099882
+      "required_token_positions_per_second": 75232.062145887,
+      "required_pages_per_second_upper_estimate": 0.0065209244728872025
     },
     {
       "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": 9411.15449919814,
-      "required_pages_per_second_upper_estimate": 0.0037542502390290966
+      "required_token_positions_per_second": 9404.007768235875,
+      "required_pages_per_second_upper_estimate": 0.0037513993011950996
     },
     {
       "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": 18822.30899839628,
-      "required_pages_per_second_upper_estimate": 0.004354209925454587
+      "required_token_positions_per_second": 18808.01553647175,
+      "required_pages_per_second_upper_estimate": 0.004350903384594685
     },
     {
       "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": 28233.463497594417,
-      "required_pages_per_second_upper_estimate": 0.0017192697974685819
+      "required_token_positions_per_second": 28212.023304707625,
+      "required_pages_per_second_upper_estimate": 0.0017179642022097747
     },
     {
       "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": 28233.463497594417,
-      "required_pages_per_second_upper_estimate": 0.002173490175273937
+      "required_token_positions_per_second": 28212.023304707625,
+      "required_pages_per_second_upper_estimate": 0.002171839649875259
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28233.463497594417,
-      "required_pages_per_second_upper_estimate": 0.3505651253162449
+      "required_token_positions_per_second": 28212.023304707625,
+      "required_pages_per_second_upper_estimate": 0.35029890987630063
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09746277219517196,
-    "producer_p95_seconds": 0.2983498219982721,
-    "gpu_update_seconds": 0.9792634900197668,
-    "producer_batches_per_second": 10.260327892146057,
-    "gpu_batches_per_second": 1.0211756184025758,
-    "producer_headroom": 10.047564500410104,
+    "producer_batch_seconds": 0.1118034327186308,
+    "producer_p95_seconds": 0.3417166580002231,
+    "gpu_update_seconds": 0.9800076974765043,
+    "producer_batches_per_second": 8.944269202508673,
+    "gpu_batches_per_second": 1.020400148463094,
+    "producer_headroom": 8.765452666760535,
     "sustainable": true,
-    "buffer_seconds": 60.714336381225536,
+    "buffer_seconds": 60.760477243543264,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368183.7058667921,
-    "upload_bytes_per_second": 4121481.18313686,
+    "generation_bytes_per_second": 366243.12396517163,
+    "upload_bytes_per_second": 4052875.5698660137,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.00365240099738,
+    "estimated_full_shards_per_hour": 78.58724869934427,
     "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": 1.6930653950003034,
-      "bytes": 12802437,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.2719050730011077,
       "bytes": 5991225,
@@ -562,6 +556,12 @@
       "bytes": 6176898,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9380000849996577,
+      "bytes": 2661456,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -209,3 +209,4 @@
 {"step": 105900, "loss": 0.6305570360273123, "lr": 0.0003, "grad_norm": 0.13164483919808131, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15496873678, "tokens_per_second": 136118.9266714003, "data_wait_s": 1.3787997886538506e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 106000, "loss": 0.6038626618683338, "lr": 0.0003, "grad_norm": 0.13520886709246546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15511593792, "tokens_per_second": 149484.15233918777, "data_wait_s": 1.5450998034793884e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 106100, "loss": 0.6192547734826803, "lr": 0.0003, "grad_norm": 0.1329948030528072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15526298252, "tokens_per_second": 149190.17782175785, "data_wait_s": 1.518700082669966e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 106200, "loss": 0.6398537121713161, "lr": 0.0003, "grad_norm": 0.14657011889388213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15540802042, "tokens_per_second": 146805.74619366202, "data_wait_s": 1.0159998055314645e-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 sha256:76adebf89f2d1df0a39f8066bade20a1dfa049dbd2428931e0388f52e64b60b1
-size 450195
+oid sha256:fcf512ba30b3d1b404acb83f197c547d3ef4ca6046cd2d5dae3033dee7230422
+size 452313