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": 74761.3673658862,
-      "required_pages_per_second_upper_estimate": 0.006480125842321746
+      "required_token_positions_per_second": 73751.19464864311,
+      "required_pages_per_second_upper_estimate": 0.00639256663145046
     },
     {
       "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": 9345.170920735774,
-      "required_pages_per_second_upper_estimate": 0.0037279284030380462
+      "required_token_positions_per_second": 9218.899331080389,
+      "required_pages_per_second_upper_estimate": 0.0036775567779960064
     },
     {
       "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": 18690.34184147155,
-      "required_pages_per_second_upper_estimate": 0.004323681646242687
+      "required_token_positions_per_second": 18437.798662160778,
+      "required_pages_per_second_upper_estimate": 0.004265260226317294
     },
     {
       "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": 28035.512762207327,
-      "required_pages_per_second_upper_estimate": 0.001707215636250752
+      "required_token_positions_per_second": 27656.697993241167,
+      "required_pages_per_second_upper_estimate": 0.0016841478043082044
     },
     {
       "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": 28035.512762207327,
-      "required_pages_per_second_upper_estimate": 0.002158251379701132
+      "required_token_positions_per_second": 27656.697993241167,
+      "required_pages_per_second_upper_estimate": 0.002129089170159721
     },
     {
       "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": 28035.512762207327,
-      "required_pages_per_second_upper_estimate": 0.3481072396812313
+      "required_token_positions_per_second": 27656.697993241167,
+      "required_pages_per_second_upper_estimate": 0.3434036280621474
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.12026655354674176,
-    "producer_p95_seconds": 0.39617068100051256,
-    "gpu_update_seconds": 0.9861777893811272,
-    "producer_batches_per_second": 8.31486369659166,
-    "gpu_batches_per_second": 1.0140159419201145,
-    "producer_headroom": 8.199933899310151,
+    "producer_batch_seconds": 0.13193945856232858,
+    "producer_p95_seconds": 0.4011740239948267,
+    "gpu_update_seconds": 0.9996855013840302,
+    "producer_batches_per_second": 7.579233770522085,
+    "gpu_batches_per_second": 1.0003145975564658,
+    "producer_headroom": 7.576850111991144,
     "sustainable": true,
-    "buffer_seconds": 60.15684515224876,
+    "buffer_seconds": 60.98081558442584,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 512534.77499691834,
-    "upload_bytes_per_second": 5626025.5484402925,
+    "generation_bytes_per_second": 510877.04183250933,
+    "upload_bytes_per_second": 5679855.998438208,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.97803151541387,
+    "estimated_full_shards_per_hour": 109.62232056838474,
     "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": 1.0173681279993616,
-      "bytes": 2612410,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2024259139980131,
       "bytes": 11156038,
@@ -562,6 +556,12 @@
       "bytes": 4790673,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.9963719520019367,
+      "bytes": 6449131,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -394,3 +394,4 @@
 {"step": 394400, "loss": 0.5831971168518066, "lr": 0.0003, "grad_norm": 0.14931650550310513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57728767715, "tokens_per_second": 149608.82075279692, "data_wait_s": 9.424999007023871e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 394500, "loss": 0.5658972728997469, "lr": 0.0003, "grad_norm": 0.17508379538776234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57743429036, "tokens_per_second": 147992.25440971044, "data_wait_s": 1.060099748428911e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 394600, "loss": 0.6006249086931348, "lr": 0.0003, "grad_norm": 0.16905402862110158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57757866352, "tokens_per_second": 146327.3491238891, "data_wait_s": 1.1219999578315765e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 394700, "loss": 0.5499810837209225, "lr": 0.0003, "grad_norm": 0.1470346524536101, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57772545742, "tokens_per_second": 149039.21257245296, "data_wait_s": 9.467999916523695e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d790b4de3b82300589583bcd8e6b1120f71766396c280ee97b0717b8048d105
-size 844374
+oid sha256:8ce8d8b340cfcaaca31bc793e044b3e0c4091f293931f26a496cc33f088a3c5c
+size 846492