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": 76258.97730875321,
-      "required_pages_per_second_upper_estimate": 0.0066099348765117595
+      "required_token_positions_per_second": 76242.81228325659,
+      "required_pages_per_second_upper_estimate": 0.006608533733071595
     },
     {
       "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": 9532.372163594151,
-      "required_pages_per_second_upper_estimate": 0.003802605777722256
+      "required_token_positions_per_second": 9530.351535407073,
+      "required_pages_per_second_upper_estimate": 0.0038017997189273467
     },
     {
       "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": 19064.744327188302,
-      "required_pages_per_second_upper_estimate": 0.004410293072054561
+      "required_token_positions_per_second": 19060.703070814146,
+      "required_pages_per_second_upper_estimate": 0.004409358198516082
     },
     {
       "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": 28597.116490782453,
-      "required_pages_per_second_upper_estimate": 0.0017414143568139312
+      "required_token_positions_per_second": 28591.05460622122,
+      "required_pages_per_second_upper_estimate": 0.0017410452198484026
     },
     {
       "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": 28597.116490782453,
-      "required_pages_per_second_upper_estimate": 0.002201485189345466
+      "required_token_positions_per_second": 28591.05460622122,
+      "required_pages_per_second_upper_estimate": 0.0022010185286915718
     },
     {
       "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": 28597.116490782453,
-      "required_pages_per_second_upper_estimate": 0.3550804784233638
+      "required_token_positions_per_second": 28591.05460622122,
+      "required_pages_per_second_upper_estimate": 0.35500521010493585
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10864058887523242,
-    "producer_p95_seconds": 0.33304966099967714,
-    "gpu_update_seconds": 0.9668107625085776,
-    "producer_batches_per_second": 9.204662919753163,
-    "gpu_batches_per_second": 1.0343285767788792,
-    "producer_headroom": 8.899167176080987,
+    "producer_batch_seconds": 0.42934188478130864,
+    "producer_p95_seconds": 0.35306783400301356,
+    "gpu_update_seconds": 0.9670157460363138,
+    "producer_batches_per_second": 2.329146154723209,
+    "gpu_batches_per_second": 1.0341093245884412,
+    "producer_headroom": 2.2523210064372754,
     "sustainable": true,
-    "buffer_seconds": 60.909078038040384,
+    "buffer_seconds": 60.921992000287766,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 462967.1187985985,
-    "upload_bytes_per_second": 4551169.508737466,
+    "generation_bytes_per_second": 409379.73317707435,
+    "upload_bytes_per_second": 4521622.285541046,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.34196637123551,
+    "estimated_full_shards_per_hour": 87.84336086734936,
     "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": 1.705037297000672,
-      "bytes": 7666167,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.7579026999992493,
       "bytes": 928826,
@@ -562,6 +556,12 @@
       "bytes": 6293747,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1847501439988264,
+      "bytes": 2572744,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -325,3 +325,4 @@
 {"step": 187500, "loss": 0.6721691600978374, "lr": 0.0003, "grad_norm": 0.17449702625152358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27441551250, "tokens_per_second": 148510.3078524755, "data_wait_s": 1.0808005754370242e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 187600, "loss": 0.569325465708971, "lr": 0.0003, "grad_norm": 0.14907051043626307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27456177803, "tokens_per_second": 151009.2038367556, "data_wait_s": 6.381997081916779e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 187700, "loss": 0.5929309464991093, "lr": 0.0003, "grad_norm": 0.14143837608252244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27471036553, "tokens_per_second": 152747.38312879467, "data_wait_s": 7.114002073649317e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 187800, "loss": 0.6040479615330696, "lr": 0.0003, "grad_norm": 0.14987953626946238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27485684795, "tokens_per_second": 149245.0930823913, "data_wait_s": 7.321999873965979e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6136c0525c76a4835590056d7f063d3489ea11df456db34461c60a32f2cfaf9c
-size 697449
+oid sha256:3862e63681150428545bb76d360d3048a064bcddb02b62afe2d4399ca039c5a1
+size 699567