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": 75625.78202531206,
-      "required_pages_per_second_upper_estimate": 0.0065550511141618546
+      "required_token_positions_per_second": 76117.96715152405,
+      "required_pages_per_second_upper_estimate": 0.0065977124734701124
     },
     {
       "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": 9453.222753164007,
-      "required_pages_per_second_upper_estimate": 0.003771031894512529
+      "required_token_positions_per_second": 9514.745893940506,
+      "required_pages_per_second_upper_estimate": 0.0037955743952212007
     },
     {
       "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": 18906.445506328015,
-      "required_pages_per_second_upper_estimate": 0.004373673425812628
+      "required_token_positions_per_second": 19029.491787881012,
+      "required_pages_per_second_upper_estimate": 0.004402138017509321
     },
     {
       "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": 28359.668259492024,
-      "required_pages_per_second_upper_estimate": 0.0017269550053229336
+      "required_token_positions_per_second": 28544.237681821516,
+      "required_pages_per_second_upper_estimate": 0.0017381943147818796
     },
     {
       "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": 28359.668259492024,
-      "required_pages_per_second_upper_estimate": 0.0021832057672019553
+      "required_token_positions_per_second": 28544.237681821516,
+      "required_pages_per_second_upper_estimate": 0.002197414432253741
     },
     {
       "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": 28359.668259492024,
-      "required_pages_per_second_upper_estimate": 0.35213216607884607
+      "required_token_positions_per_second": 28544.237681821516,
+      "required_pages_per_second_upper_estimate": 0.3544239005900582
     }
   ],
   "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": 0.4977076557030671,
-    "producer_p95_seconds": 0.34365429199897335,
-    "gpu_update_seconds": 0.9749056211454862,
-    "producer_batches_per_second": 2.009211609548962,
-    "gpu_batches_per_second": 1.0257403160985252,
-    "producer_headroom": 1.9587916922200528,
+    "producer_batch_seconds": 0.12323137025782671,
+    "producer_p95_seconds": 0.34655145299984724,
+    "gpu_update_seconds": 0.9686018000616535,
+    "producer_batches_per_second": 8.114816851486626,
+    "gpu_batches_per_second": 1.0324160041168082,
+    "producer_headroom": 7.860026209520585,
     "sustainable": true,
-    "buffer_seconds": 60.444148511020146,
+    "buffer_seconds": 60.05331160382252,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276226.06679244776,
-    "upload_bytes_per_second": 3680972.5188452546,
+    "generation_bytes_per_second": 276341.4114508037,
+    "upload_bytes_per_second": 3691439.0666800505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.271683719921825,
+    "estimated_full_shards_per_hour": 59.29643399851878,
     "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": 1.0122808699998131,
-      "bytes": 6141256,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8550289739978325,
       "bytes": 5835711,
@@ -562,6 +556,12 @@
       "bytes": 5543356,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 0.995166409000376,
+      "bytes": 7263909,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -303,3 +303,4 @@
 {"step": 70300, "loss": 0.6057273186743259, "lr": 0.0003, "grad_norm": 0.1319615624256686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10288636575, "tokens_per_second": 149223.92229449312, "data_wait_s": 8.087998139671981e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 70400, "loss": 0.6433261074125767, "lr": 0.0003, "grad_norm": 0.13315337529539414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10303344006, "tokens_per_second": 130383.0157648406, "data_wait_s": 1.4115001249592751e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 70500, "loss": 0.6558354012668133, "lr": 0.0003, "grad_norm": 0.13466285050920596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10317975409, "tokens_per_second": 149755.84670505646, "data_wait_s": 1.0271000064676628e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 70600, "loss": 0.5950672663748264, "lr": 0.0003, "grad_norm": 0.13279087819561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10332681970, "tokens_per_second": 128305.54160269313, "data_wait_s": 1.6959998902166262e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c9423d8e8ad109eb3432e96f5d1b36586cff5e98bb13c1c429d49c7bf9e17eeb
-size 650853
+oid sha256:52b22fe2ae8465edfb1bc7e85bf4b016851500f80124455022ee8c5479b99e0d
+size 652971