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": 75573.97470236004,
-      "required_pages_per_second_upper_estimate": 0.006550560586712833
+      "required_token_positions_per_second": 74962.50811679951,
+      "required_pages_per_second_upper_estimate": 0.006497560212824339
     },
     {
       "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": 9446.746837795004,
-      "required_pages_per_second_upper_estimate": 0.0037684485550482706
+      "required_token_positions_per_second": 9370.313514599939,
+      "required_pages_per_second_upper_estimate": 0.003737958159645739
     },
     {
       "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": 18893.49367559001,
-      "required_pages_per_second_upper_estimate": 0.004370677247715826
+      "required_token_positions_per_second": 18740.627029199877,
+      "required_pages_per_second_upper_estimate": 0.004335314239434558
     },
     {
       "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": 28340.24051338501,
-      "required_pages_per_second_upper_estimate": 0.0017257719575145228
+      "required_token_positions_per_second": 28110.940543799814,
+      "required_pages_per_second_upper_estimate": 0.0017118087924107486
     },
     {
       "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": 28340.24051338501,
-      "required_pages_per_second_upper_estimate": 0.0021817101655271943
+      "required_token_positions_per_second": 28110.940543799814,
+      "required_pages_per_second_upper_estimate": 0.0021640580191255848
     },
     {
       "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": 28340.24051338501,
-      "required_pages_per_second_upper_estimate": 0.35189093849268055
+      "required_token_positions_per_second": 28110.940543799814,
+      "required_pages_per_second_upper_estimate": 0.34904380028806403
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2937858406563123,
-    "producer_p95_seconds": 0.32033154900273075,
-    "gpu_update_seconds": 0.9755739365352927,
-    "producer_batches_per_second": 3.40384001409332,
-    "gpu_batches_per_second": 1.0250376343093537,
-    "producer_headroom": 3.3206976018853664,
+    "producer_batch_seconds": 0.13454121862514512,
+    "producer_p95_seconds": 0.3355276980000781,
+    "gpu_update_seconds": 0.9835316593879702,
+    "producer_batches_per_second": 7.432666436493126,
+    "gpu_batches_per_second": 1.0167440879557226,
+    "producer_headroom": 7.310262753961355,
     "sustainable": true,
-    "buffer_seconds": 60.48558406518815,
+    "buffer_seconds": 60.97896288205415,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 461112.9337084278,
-    "upload_bytes_per_second": 2667768.6443245206,
+    "generation_bytes_per_second": 443163.14938359824,
+    "upload_bytes_per_second": 2724785.6344628916,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.94410141410471,
+    "estimated_full_shards_per_hour": 95.09249554758988,
     "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": 0.8424355740025931,
-      "bytes": 669079,
-      "pages": 20,
-      "shards": 1
-    },
     {
       "seconds": 1.4896883629990043,
       "bytes": 13505965,
@@ -562,6 +556,12 @@
       "bytes": 13551063,
       "pages": 43,
       "shards": 4
+    },
+    {
+      "seconds": 1.137746813998092,
+      "bytes": 10438351,
+      "pages": 27,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -216,3 +216,4 @@
 {"step": 301800, "loss": 0.5959030166268349, "lr": 0.0003, "grad_norm": 0.1757241827194646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44175758587, "tokens_per_second": 151618.25286244875, "data_wait_s": 9.413997759111226e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 301900, "loss": 0.5809973161667585, "lr": 0.0003, "grad_norm": 0.1742551423624091, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44190476385, "tokens_per_second": 150873.22175957926, "data_wait_s": 8.486997103318572e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 302000, "loss": 0.596163921058178, "lr": 0.0003, "grad_norm": 0.1650319641062203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44205122020, "tokens_per_second": 149900.3523269715, "data_wait_s": 1.003199940896593e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 302100, "loss": 0.663341885432601, "lr": 0.0003, "grad_norm": 0.19818130432907602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44219840479, "tokens_per_second": 150284.3415450765, "data_wait_s": 1.247799809789285e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2664c77d92535e25fbea6f1d15a2e5cfc65d7485d2b7d129da29e4120ec869fb
-size 465021
+oid sha256:abc8d7d055b898664942597ed1dbe68ca59287577235f29478782dc1b7fffeb4
+size 467139