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": 75097.39572577628,
-      "required_pages_per_second_upper_estimate": 0.006509251928900926
+      "required_token_positions_per_second": 74962.45005463828,
+      "required_pages_per_second_upper_estimate": 0.006497555180143361
     },
     {
       "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": 9387.174465722035,
-      "required_pages_per_second_upper_estimate": 0.0037446842451420276
+      "required_token_positions_per_second": 9370.306256829785,
+      "required_pages_per_second_upper_estimate": 0.003737955264412711
     },
     {
       "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": 18774.34893144407,
-      "required_pages_per_second_upper_estimate": 0.0043431152080335425
+      "required_token_positions_per_second": 18740.61251365957,
+      "required_pages_per_second_upper_estimate": 0.004335310881519772
     },
     {
       "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": 28161.523397166104,
-      "required_pages_per_second_upper_estimate": 0.0017148890227930342
+      "required_token_positions_per_second": 28110.918770489356,
+      "required_pages_per_second_upper_estimate": 0.0017118074665303773
     },
     {
       "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": 28161.523397166104,
-      "required_pages_per_second_upper_estimate": 0.0021679520272000216
+      "required_token_positions_per_second": 28110.918770489356,
+      "required_pages_per_second_upper_estimate": 0.002164056342955873
     },
     {
       "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": 28161.523397166104,
-      "required_pages_per_second_upper_estimate": 0.34967187003695327
+      "required_token_positions_per_second": 28110.918770489356,
+      "required_pages_per_second_upper_estimate": 0.34904352993641874
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.1218912077341372,
-    "producer_p95_seconds": 0.33288223999988986,
-    "gpu_update_seconds": 0.9817650703790485,
-    "producer_batches_per_second": 8.204037178638416,
-    "gpu_batches_per_second": 1.0185736182424083,
-    "producer_headroom": 8.054437138078274,
+    "producer_batch_seconds": 0.07306469193738963,
+    "producer_p95_seconds": 0.26177931699930923,
+    "gpu_update_seconds": 0.9835324211823584,
+    "producer_batches_per_second": 13.686501283778997,
+    "gpu_batches_per_second": 1.0167433004372597,
+    "producer_headroom": 13.461117745150613,
     "sustainable": true,
-    "buffer_seconds": 60.86943436350101,
+    "buffer_seconds": 60.979010113306224,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 469204.45198427676,
-    "upload_bytes_per_second": 5773558.912988035,
+    "generation_bytes_per_second": 469712.6360687602,
+    "upload_bytes_per_second": 5740644.757336163,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.68035287519672,
+    "estimated_full_shards_per_hour": 100.78939734980682,
     "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": 0.8392134009991423,
-      "bytes": 3718175,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.0338640699992538,
       "bytes": 9654277,
@@ -562,6 +556,12 @@
       "bytes": 12108906,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 0.808165985999949,
+      "bytes": 1135299,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 21900, "loss": 0.6555679701268673, "lr": 0.0003, "grad_norm": 0.1287331654688954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3204134161, "tokens_per_second": 146275.9905816803, "data_wait_s": 9.554001735523343e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 22000, "loss": 0.6465682405978441, "lr": 0.0003, "grad_norm": 0.1496552983906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3218668632, "tokens_per_second": 147316.45437906487, "data_wait_s": 1.2208001862745732e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 22100, "loss": 0.6416172496974468, "lr": 0.0003, "grad_norm": 0.13793537535635847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3233365554, "tokens_per_second": 149169.9726239334, "data_wait_s": 1.3908997061662376e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 22200, "loss": 0.6939101740717888, "lr": 0.0003, "grad_norm": 0.1335519718057603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3248054489, "tokens_per_second": 149274.7079086694, "data_wait_s": 1.0237999958917499e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c9ad94ea00722b8719e8f8a15eb6b0277e89b5d966817c97afa5fe2233e692b
-size 465601
+oid sha256:b4e426cdc4da0f1202249a90b312b5da649c59c143b78fc5932dfe0def5a5d68
+size 467719