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": 74474.73470015317,
-      "required_pages_per_second_upper_estimate": 0.006455281249319857
+      "required_token_positions_per_second": 74444.09056524152,
+      "required_pages_per_second_upper_estimate": 0.006452625093372565
     },
     {
       "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": 9309.341837519147,
-      "required_pages_per_second_upper_estimate": 0.003713635646050402
+      "required_token_positions_per_second": 9305.51132065519,
+      "required_pages_per_second_upper_estimate": 0.0037121075956020387
     },
     {
       "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": 18618.683675038294,
-      "required_pages_per_second_upper_estimate": 0.004307104790579009
+      "required_token_positions_per_second": 18611.02264131038,
+      "required_pages_per_second_upper_estimate": 0.004305332545255646
     },
     {
       "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": 27928.025512557444,
-      "required_pages_per_second_upper_estimate": 0.0017006702266890892
+      "required_token_positions_per_second": 27916.53396196557,
+      "required_pages_per_second_upper_estimate": 0.0016999704515495504
     },
     {
       "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": 27928.025512557444,
-      "required_pages_per_second_upper_estimate": 0.002149976713679333
+      "required_token_positions_per_second": 27916.53396196557,
+      "required_pages_per_second_upper_estimate": 0.002149092062304123
     },
     {
       "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": 27928.025512557444,
-      "required_pages_per_second_upper_estimate": 0.34677260777726315
+      "required_token_positions_per_second": 27916.53396196557,
+      "required_pages_per_second_upper_estimate": 0.34662992117865793
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06711999782024236,
-    "producer_p95_seconds": 0.17337114800011477,
-    "gpu_update_seconds": 0.989973315069068,
-    "producer_batches_per_second": 14.898689399218297,
-    "gpu_batches_per_second": 1.0101282375780325,
-    "producer_headroom": 14.749304934728519,
+    "producer_batch_seconds": 0.08142173799211605,
+    "producer_p95_seconds": 0.17337984299956588,
+    "gpu_update_seconds": 0.990380827278507,
+    "producer_batches_per_second": 12.281732429941849,
+    "gpu_batches_per_second": 1.0097125998974816,
+    "producer_headroom": 12.163592324379076,
     "sustainable": true,
-    "buffer_seconds": 60.38837221921315,
+    "buffer_seconds": 60.413230463988924,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304649.2589043096,
+    "generation_bytes_per_second": 327987.96301960835,
     "upload_bytes_per_second": 5280629.927993453,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.37063908907858,
+    "estimated_full_shards_per_hour": 70.37858169499576,
     "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": [

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 319100, "loss": 0.5913281161338091, "lr": 0.0003, "grad_norm": 0.13908416392842843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46707670445, "tokens_per_second": 148139.4046867686, "data_wait_s": 1.4523999197990634e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 319200, "loss": 0.5889566745609045, "lr": 0.0003, "grad_norm": 0.15953488716539, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46722180635, "tokens_per_second": 147244.37958067923, "data_wait_s": 1.4233000001695473e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 319300, "loss": 0.5744155272841454, "lr": 0.0003, "grad_norm": 0.13654334719865105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46736730170, "tokens_per_second": 147518.4826699028, "data_wait_s": 1.6621000213490333e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 319400, "loss": 0.5801581870764494, "lr": 0.0003, "grad_norm": 0.1494298392152412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46751229030, "tokens_per_second": 146928.5534759068, "data_wait_s": 1.0874999134102836e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0ef28b814b8544ab676e68cd58fd2ba345f94a1b669ac69b3cb7bbdcfc0c1b4
-size 91239
+oid sha256:d9e6654814d25be9b08875b4e61ac570545dab8168517b9a4ddbcc10a467b49a
+size 93357