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": 75278.37641351564,
-      "required_pages_per_second_upper_estimate": 0.006524938876222814
+      "required_token_positions_per_second": 75313.04526315643,
+      "required_pages_per_second_upper_estimate": 0.006527943884242281
     },
     {
       "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": 9409.797051689455,
-      "required_pages_per_second_upper_estimate": 0.0037537087329222335
+      "required_token_positions_per_second": 9414.130657894553,
+      "required_pages_per_second_upper_estimate": 0.003755437473230634
     },
     {
       "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": 18819.59410337891,
-      "required_pages_per_second_upper_estimate": 0.004353581882272839
+      "required_token_positions_per_second": 18828.261315789106,
+      "required_pages_per_second_upper_estimate": 0.004355586889326209
     },
     {
       "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": 28229.391155068362,
-      "required_pages_per_second_upper_estimate": 0.001719021813174675
+      "required_token_positions_per_second": 28242.391973683658,
+      "required_pages_per_second_upper_estimate": 0.0017198134948183228
     },
     {
       "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": 28229.391155068362,
-      "required_pages_per_second_upper_estimate": 0.0021731766750733156
+      "required_token_positions_per_second": 28242.391973683658,
+      "required_pages_per_second_upper_estimate": 0.0021741775140788904
     },
     {
       "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": 28229.391155068362,
-      "required_pages_per_second_upper_estimate": 0.350514560451325
+      "required_token_positions_per_second": 28242.391973683658,
+      "required_pages_per_second_upper_estimate": 0.3506759871075861
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.3968630505546571,
-    "producer_p95_seconds": 0.1964224369985459,
-    "gpu_update_seconds": 0.9794047575495096,
-    "producer_batches_per_second": 2.5197609064446707,
-    "gpu_batches_per_second": 1.0210283259211648,
-    "producer_headroom": 2.4678658196591754,
+    "producer_batch_seconds": 0.07213930092177634,
+    "producer_p95_seconds": 0.18885419300022477,
+    "gpu_update_seconds": 0.9789539082157944,
+    "producer_batches_per_second": 13.862069457594853,
+    "gpu_batches_per_second": 1.0214985522889055,
+    "producer_headroom": 13.57032707147128,
     "sustainable": true,
-    "buffer_seconds": 60.723094968069596,
+    "buffer_seconds": 60.69514230937926,
     "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": 350453.91113505006,
-    "upload_bytes_per_second": 5673522.257324757,
+    "generation_bytes_per_second": 363048.9051972244,
+    "upload_bytes_per_second": 5705399.740533962,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.1992511800635,
+    "estimated_full_shards_per_hour": 77.9018437093501,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 5601597,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 1.1996037619992421,
+      "bytes": 8790333,
+      "pages": 45,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 364000, "loss": 0.5973384417593479, "lr": 0.0003, "grad_norm": 0.18057876754783023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53279484924, "tokens_per_second": 147822.25848263092, "data_wait_s": 1.3679999028681777e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 364100, "loss": 0.5626239739358425, "lr": 0.0003, "grad_norm": 0.15033332213220368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53293948478, "tokens_per_second": 136348.8640023422, "data_wait_s": 1.2354999853414483e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 364200, "loss": 0.6012831311672926, "lr": 0.0003, "grad_norm": 0.15035212828651234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53308485969, "tokens_per_second": 148375.70411016783, "data_wait_s": 1.0566000128164887e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 364300, "loss": 0.5770778506994247, "lr": 0.0003, "grad_norm": 0.15973130305345476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53323134573, "tokens_per_second": 149729.65350880558, "data_wait_s": 7.982000170159154e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f0d686b9b4a6b15808f98d20ac2e18f8464ebff7507c52541a39a1525785f6b3
-size 195804
+oid sha256:80de382f91d6454bbb3415df6d1999ba71dc1c7cc0173de193ca30ff369d404e
+size 197922