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": 76153.05674145524,
-      "required_pages_per_second_upper_estimate": 0.006600753950191595
+      "required_token_positions_per_second": 75039.92905469185,
+      "required_pages_per_second_upper_estimate": 0.006504270863499274
     },
     {
       "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": 9519.132092681904,
-      "required_pages_per_second_upper_estimate": 0.00379732411547866
+      "required_token_positions_per_second": 9379.991131836481,
+      "required_pages_per_second_upper_estimate": 0.0037418187058546677
     },
     {
       "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": 19038.26418536381,
-      "required_pages_per_second_upper_estimate": 0.004404167357278044
+      "required_token_positions_per_second": 18759.982263672962,
+      "required_pages_per_second_upper_estimate": 0.004339791732289414
     },
     {
       "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": 28557.39627804571,
-      "required_pages_per_second_upper_estimate": 0.0017389956042541143
+      "required_token_positions_per_second": 28139.97339550944,
+      "required_pages_per_second_upper_estimate": 0.0017135767407562641
     },
     {
       "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": 28557.39627804571,
-      "required_pages_per_second_upper_estimate": 0.0021984274174164055
+      "required_token_positions_per_second": 28139.97339550944,
+      "required_pages_per_second_upper_estimate": 0.002166293048418269
     },
     {
       "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": 28557.39627804571,
-      "required_pages_per_second_upper_estimate": 0.3545872863161741
+      "required_token_positions_per_second": 28139.97339550944,
+      "required_pages_per_second_upper_estimate": 0.3494042911395935
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09626990489057619,
-    "producer_p95_seconds": 0.1877291119999427,
-    "gpu_update_seconds": 0.9681554904658856,
-    "producer_batches_per_second": 10.387462220271598,
-    "gpu_batches_per_second": 1.0328919371399634,
-    "producer_headroom": 10.056678580562904,
+    "producer_batch_seconds": 0.05663709317181542,
+    "producer_p95_seconds": 0.1854400149986759,
+    "gpu_update_seconds": 0.9825169203753422,
+    "producer_batches_per_second": 17.656273371346582,
+    "gpu_batches_per_second": 1.0177941766315626,
+    "producer_headroom": 17.347587338120604,
     "sustainable": true,
-    "buffer_seconds": 60.025640408884904,
+    "buffer_seconds": 60.91604906327122,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443971.94246196677,
-    "upload_bytes_per_second": 5968194.182742622,
+    "generation_bytes_per_second": 394326.17851746845,
+    "upload_bytes_per_second": 5928699.59931897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.2660437144685,
+    "estimated_full_shards_per_hour": 84.61321846621551,
     "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": 7569670,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.248841076001554,
+      "bytes": 5066284,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -90,3 +90,4 @@
 {"step": 49000, "loss": 0.5643719397485256, "lr": 0.0003, "grad_norm": 0.12859644528598635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7170937232, "tokens_per_second": 149682.37203272746, "data_wait_s": 9.879000572254881e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 49100, "loss": 0.6665434390306473, "lr": 0.0003, "grad_norm": 0.1404694792377278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7185626821, "tokens_per_second": 151023.4391343018, "data_wait_s": 7.995000487426296e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 49200, "loss": 0.7101405002176762, "lr": 0.0003, "grad_norm": 0.1457871580642793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7200327105, "tokens_per_second": 143859.85854952806, "data_wait_s": 1.0122999810846522e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 49300, "loss": 0.6631663143634796, "lr": 0.0003, "grad_norm": 0.1397440728156991, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7215087862, "tokens_per_second": 151228.84642449685, "data_wait_s": 7.808999725966714e-06, "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:f000a4f68cd9cecf68554ecd950eeee4e602158a0ab327001c25ead0995671a3
-size 195804
+oid sha256:996ecfdffecb2fdcd37d8248077568efbffada849a94ef58004c7cf738610cba
+size 197922