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": 75322.99995135148,
-      "required_pages_per_second_upper_estimate": 0.0065288067313851025
+      "required_token_positions_per_second": 75463.75289427815,
+      "required_pages_per_second_upper_estimate": 0.0065410068397429135
     },
     {
       "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": 9415.374993918935,
-      "required_pages_per_second_upper_estimate": 0.0037559338574752413
+      "required_token_positions_per_second": 9432.969111784769,
+      "required_pages_per_second_upper_estimate": 0.0037629524141474264
     },
     {
       "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": 18830.74998783787,
-      "required_pages_per_second_upper_estimate": 0.004356162599805558
+      "required_token_positions_per_second": 18865.938223569538,
+      "required_pages_per_second_upper_estimate": 0.004364302778850282
     },
     {
       "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": 28246.124981756802,
-      "required_pages_per_second_upper_estimate": 0.0017200408154243976
+      "required_token_positions_per_second": 28298.907335354306,
+      "required_pages_per_second_upper_estimate": 0.0017232549838308783
     },
     {
       "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": 28246.124981756802,
-      "required_pages_per_second_upper_estimate": 0.0021744648913739886
+      "required_token_positions_per_second": 28298.907335354306,
+      "required_pages_per_second_upper_estimate": 0.002178528223064831
     },
     {
       "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": 28246.124981756802,
-      "required_pages_per_second_upper_estimate": 0.35072233857428015
+      "required_token_positions_per_second": 28298.907335354306,
+      "required_pages_per_second_upper_estimate": 0.3513777187547873
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.07613497392969748,
-    "producer_p95_seconds": 0.19605788999979268,
-    "gpu_update_seconds": 0.978824529660507,
-    "producer_batches_per_second": 13.134568101689943,
-    "gpu_batches_per_second": 1.0216335713887732,
-    "producer_headroom": 12.856437444430558,
+    "producer_batch_seconds": 0.053026351343760325,
+    "producer_p95_seconds": 0.1887002740004391,
+    "gpu_update_seconds": 0.9769988527245673,
+    "producer_batches_per_second": 18.858548149337665,
+    "gpu_batches_per_second": 1.0235426553585905,
+    "producer_headroom": 18.42477990595391,
     "sustainable": true,
-    "buffer_seconds": 60.68712083895144,
+    "buffer_seconds": 60.573928868923176,
     "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": 347732.56317525177,
-    "upload_bytes_per_second": 6006490.58822615,
+    "generation_bytes_per_second": 354473.1449605094,
+    "upload_bytes_per_second": 5810467.591875715,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.61531325762905,
+    "estimated_full_shards_per_hour": 76.06168519603216,
     "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": [
@@ -280,6 +280,12 @@
       "bytes": 13322526,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.8405508110008668,
+      "bytes": 1090490,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -20,3 +20,4 @@
 {"step": 282200, "loss": 0.6118101850152016, "lr": 0.0003, "grad_norm": 0.16666576891706478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41306225341, "tokens_per_second": 147814.78691748617, "data_wait_s": 9.528999726171605e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 282300, "loss": 0.6148794535547495, "lr": 0.0003, "grad_norm": 0.16453517197819229, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41320742235, "tokens_per_second": 148631.58136930884, "data_wait_s": 1.3801998647977598e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 282400, "loss": 0.5748421065509319, "lr": 0.0003, "grad_norm": 0.15926820594003518, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41335378402, "tokens_per_second": 149925.13206825987, "data_wait_s": 1.2630000128410757e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 282500, "loss": 0.583519360050559, "lr": 0.0003, "grad_norm": 0.1592176146246623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41350063623, "tokens_per_second": 149840.68864331825, "data_wait_s": 9.033999958774075e-06, "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 sha256:2f0d9595a645700536a39957f26a1d3879ec31b90d4b48edee553a8919412e79
-size 46761
+oid sha256:fb24c4b877a2ef5dc08452c6f789bb4dfd68be3f18b0856d1c53ed38c6a6b8bc
+size 48879