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": 74600.37100032733,
-      "required_pages_per_second_upper_estimate": 0.006466171085396661
+      "required_token_positions_per_second": 74694.80277680648,
+      "required_pages_per_second_upper_estimate": 0.006474356192446724
     },
     {
       "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": 9325.046375040916,
-      "required_pages_per_second_upper_estimate": 0.0037199004208715958
+      "required_token_positions_per_second": 9336.85034710081,
+      "required_pages_per_second_upper_estimate": 0.003724609201811397
     },
     {
       "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": 18650.092750081832,
-      "required_pages_per_second_upper_estimate": 0.004314370727309492
+      "required_token_positions_per_second": 18673.70069420162,
+      "required_pages_per_second_upper_estimate": 0.004319832009695978
     },
     {
       "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": 27975.139125122747,
-      "required_pages_per_second_upper_estimate": 0.0017035391985082944
+      "required_token_positions_per_second": 28010.55104130243,
+      "required_pages_per_second_upper_estimate": 0.001705695598411672
     },
     {
       "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": 27975.139125122747,
-      "required_pages_per_second_upper_estimate": 0.002153603649993437
+      "required_token_positions_per_second": 28010.55104130243,
+      "required_pages_per_second_upper_estimate": 0.002156329757327407
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27975.139125122747,
-      "required_pages_per_second_upper_estimate": 0.3473576011662062
+      "required_token_positions_per_second": 28010.55104130243,
+      "required_pages_per_second_upper_estimate": 0.3477972986490983
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.25633829831252797,
-    "producer_p95_seconds": 0.3414044359997206,
-    "gpu_update_seconds": 0.988306076918525,
-    "producer_batches_per_second": 3.9010947899045454,
-    "gpu_batches_per_second": 1.0118322889584326,
-    "producer_headroom": 3.855475687497859,
+    "producer_batch_seconds": 0.12232675290596262,
+    "producer_p95_seconds": 0.32659740099916235,
+    "gpu_update_seconds": 0.9870566258847304,
+    "producer_batches_per_second": 8.17482665274978,
+    "gpu_batches_per_second": 1.0131131018989594,
+    "producer_headroom": 8.069016813055764,
     "sustainable": true,
-    "buffer_seconds": 60.28667069203003,
+    "buffer_seconds": 60.210454178968554,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346120.8382687419,
-    "upload_bytes_per_second": 4115146.977753103,
+    "generation_bytes_per_second": 346262.46584288386,
+    "upload_bytes_per_second": 4154548.9287369032,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.26947461172765,
+    "estimated_full_shards_per_hour": 74.29986459221732,
     "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": 1.035596928002633,
-      "bytes": 2562080,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3296433830000751,
       "bytes": 11732086,
@@ -562,6 +556,12 @@
       "bytes": 11753109,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.0484200940009032,
+      "bytes": 6644677,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -216,3 +216,4 @@
 {"step": 106600, "loss": 0.5847213715314865, "lr": 0.0003, "grad_norm": 0.13213030274642967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15599028513, "tokens_per_second": 148926.07728392465, "data_wait_s": 1.3206001312937587e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 106700, "loss": 0.6291310619562864, "lr": 0.0003, "grad_norm": 0.139231069345712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15613778785, "tokens_per_second": 148545.86893485233, "data_wait_s": 1.749500006553717e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 106800, "loss": 0.653808118775487, "lr": 0.0003, "grad_norm": 0.12560414814596255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15628534506, "tokens_per_second": 149153.49299840332, "data_wait_s": 1.3376000424614176e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 106900, "loss": 0.6316846124827862, "lr": 0.0003, "grad_norm": 0.1368452450334674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15643086303, "tokens_per_second": 147602.88098039146, "data_wait_s": 1.6486003005411476e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c42247d68d25eee5e92b1d6bc0ea4eb2e41405888f02a7f02e58130e320f8f35
-size 465021
+oid sha256:816e09dbe20d3546b5e3ac0e6fc18a0499c00ef45f12df640d7474d908db119c
+size 467139