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": 76615.83618416266,
-      "required_pages_per_second_upper_estimate": 0.0053004613204526915
+      "required_token_positions_per_second": 76902.52100242539,
+      "required_pages_per_second_upper_estimate": 0.005320294841380533
     },
     {
       "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": 9576.979523020333,
-      "required_pages_per_second_upper_estimate": 0.004009681336214543
+      "required_token_positions_per_second": 9612.815125303174,
+      "required_pages_per_second_upper_estimate": 0.004024684954557898
     },
     {
       "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": 19153.959046040665,
-      "required_pages_per_second_upper_estimate": 0.003989264545422117
+      "required_token_positions_per_second": 19225.630250606348,
+      "required_pages_per_second_upper_estimate": 0.004004191767236381
     },
     {
       "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": 28730.938569061,
-      "required_pages_per_second_upper_estimate": 0.0017223758976121445
+      "required_token_positions_per_second": 28838.445375909523,
+      "required_pages_per_second_upper_estimate": 0.001728820766529324
     },
     {
       "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": 28730.938569061,
-      "required_pages_per_second_upper_estimate": 0.002233704099701348
+      "required_token_positions_per_second": 28838.445375909523,
+      "required_pages_per_second_upper_estimate": 0.002242062281061352
     },
     {
       "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": 28730.938569061,
-      "required_pages_per_second_upper_estimate": 0.3685296310855556
+      "required_token_positions_per_second": 28838.445375909523,
+      "required_pages_per_second_upper_estimate": 0.3699086129719927
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.4805450669140612,
-    "producer_p95_seconds": 0.24100237399761681,
-    "gpu_update_seconds": 0.962307581200039,
-    "producer_batches_per_second": 2.080970274904177,
-    "gpu_batches_per_second": 1.0391687850499494,
-    "producer_headroom": 2.0025334717922187,
+    "producer_batch_seconds": 0.44178182867986493,
+    "producer_p95_seconds": 0.24356246900060796,
+    "gpu_update_seconds": 0.958720195891559,
+    "producer_batches_per_second": 2.263560733107122,
+    "gpu_batches_per_second": 1.0430571967559867,
+    "producer_headroom": 2.170121389456901,
     "sustainable": true,
-    "buffer_seconds": 60.62537761560246,
+    "buffer_seconds": 60.39937234116822,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286533.3983847403,
-    "upload_bytes_per_second": 5029963.500237349,
+    "generation_bytes_per_second": 261696.817864385,
+    "upload_bytes_per_second": 5048326.035628393,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.48339713722855,
+    "estimated_full_shards_per_hour": 56.15404512356436,
     "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": 1.0937011059995712,
-      "bytes": 11191807,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.0167957749999914,
       "bytes": 2251564,
@@ -562,6 +556,12 @@
       "bytes": 1212322,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.1948913660016842,
+      "bytes": 13257800,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -205,3 +205,4 @@
 {"step": 460700, "loss": 0.598781168460846, "lr": 0.0003, "grad_norm": 0.15169242466307828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67432477745, "tokens_per_second": 153328.0984045403, "data_wait_s": 1.5421999705722556e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 460800, "loss": 0.5516125559806824, "lr": 0.0003, "grad_norm": 0.14995642532327907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67447147514, "tokens_per_second": 152562.41799135934, "data_wait_s": 8.582999726058915e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 460900, "loss": 0.5935230255126953, "lr": 0.0003, "grad_norm": 0.15514092447640287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67461862382, "tokens_per_second": 152511.22567640184, "data_wait_s": 9.760999091668054e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 461000, "loss": 0.5812311172485352, "lr": 0.0003, "grad_norm": 0.15962488188450133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67476569867, "tokens_per_second": 152967.32783971506, "data_wait_s": 1.3201999536249787e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebc3df665cc80b7a1624ab7a129e9037116ea93902792bc2cb01b72f78f163f2
-size 441723
+oid sha256:96f34682a59934cd3935db2f468d034aacc1b7ba0652541d24798cf20c95c4ae
+size 443841