Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76429.50861659455,
-      "required_pages_per_second_upper_estimate": 0.005287570746988809
+      "required_token_positions_per_second": 76326.47294775462,
+      "required_pages_per_second_upper_estimate": 0.005280442500342772
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9553.68857707432,
-      "required_pages_per_second_upper_estimate": 0.003999929903517206
+      "required_token_positions_per_second": 9540.809118469328,
+      "required_pages_per_second_upper_estimate": 0.003994537543152975
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19107.37715414864,
-      "required_pages_per_second_upper_estimate": 0.003979562765788107
+      "required_token_positions_per_second": 19081.618236938655,
+      "required_pages_per_second_upper_estimate": 0.003974197862641596
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28661.065731222956,
-      "required_pages_per_second_upper_estimate": 0.0017181871276733996
+      "required_token_positions_per_second": 28622.42735540798,
+      "required_pages_per_second_upper_estimate": 0.0017158708160406756
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28661.065731222956,
-      "required_pages_per_second_upper_estimate": 0.002228271793897573
+      "required_token_positions_per_second": 28622.42735540798,
+      "required_pages_per_second_upper_estimate": 0.0022252678301302132
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28661.065731222956,
-      "required_pages_per_second_upper_estimate": 0.36763337734537727
+      "required_token_positions_per_second": 28622.42735540798,
+      "required_pages_per_second_upper_estimate": 0.3671377657470784
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11538068181249628,
-    "producer_p95_seconds": 0.34181169400108047,
-    "gpu_update_seconds": 0.964653591714863,
-    "producer_batches_per_second": 8.666962131711855,
-    "gpu_batches_per_second": 1.0366415556721267,
-    "producer_headroom": 8.360616149612547,
+    "producer_batch_seconds": 0.12011707118739423,
+    "producer_p95_seconds": 0.34971746699739015,
+    "gpu_update_seconds": 0.9659558099909415,
+    "producer_batches_per_second": 8.325211313551788,
+    "gpu_batches_per_second": 1.0352440449728002,
+    "producer_headroom": 8.041786237727669,
     "sustainable": true,
-    "buffer_seconds": 60.77317627803637,
+    "buffer_seconds": 60.855216029429315,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372617.84420548286,
-    "upload_bytes_per_second": 5121692.083898635,
+    "generation_bytes_per_second": 357632.4993693951,
+    "upload_bytes_per_second": 5003102.433527964,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.95511526702275,
+    "estimated_full_shards_per_hour": 76.73960910617247,
     "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": 1.5466363100022136,
-      "bytes": 15617223,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.0353363290014386,
       "bytes": 1240846,
@@ -562,6 +556,12 @@
       "bytes": 11617672,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.6245845399971586,
+      "bytes": 5983952,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -405,3 +405,4 @@
 {"step": 965500, "loss": 0.49153828620910645, "lr": 3e-05, "grad_norm": 0.341800847071638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141322690087, "tokens_per_second": 127163.39158951252, "data_wait_s": 1.6079997294582427e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 965600, "loss": 0.5570799708366394, "lr": 3e-05, "grad_norm": 0.3516265639521352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141337341470, "tokens_per_second": 151800.38202151313, "data_wait_s": 9.75099828792736e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 965700, "loss": 0.5379848480224609, "lr": 3e-05, "grad_norm": 0.352873060302894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141351874280, "tokens_per_second": 151219.754145978, "data_wait_s": 1.1971998901572078e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 965800, "loss": 0.5314061045646667, "lr": 3e-05, "grad_norm": 0.36432927064700327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141366491370, "tokens_per_second": 152743.25672000687, "data_wait_s": 1.0431998816784471e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9dc497a3c2d7c24a37012125a53089de560b00752f23c4e969efac6cda6f0d56
-size 868455
+oid sha256:2ddbc2de49c93a6dac2b6e298a7e0a9e5de5b3c4473915097f3b052751cf71c5
+size 870573