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": 75703.71619841477,
-      "required_pages_per_second_upper_estimate": 0.006561806250764041
+      "required_token_positions_per_second": 74902.19282132518,
+      "required_pages_per_second_upper_estimate": 0.006492332235880343
     },
     {
       "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": 9462.964524801846,
-      "required_pages_per_second_upper_estimate": 0.0037749180328713284
+      "required_token_positions_per_second": 9362.774102665648,
+      "required_pages_per_second_upper_estimate": 0.003734950575500897
     },
     {
       "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": 18925.929049603692,
-      "required_pages_per_second_upper_estimate": 0.004378180600650806
+      "required_token_positions_per_second": 18725.548205331295,
+      "required_pages_per_second_upper_estimate": 0.004331826018910798
     },
     {
       "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": 28388.893574405538,
-      "required_pages_per_second_upper_estimate": 0.0017287346736677895
+      "required_token_positions_per_second": 28088.322307996947,
+      "required_pages_per_second_upper_estimate": 0.0017104314605190661
     },
     {
       "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": 28388.893574405538,
-      "required_pages_per_second_upper_estimate": 0.002185455612844847
+      "required_token_positions_per_second": 28088.322307996947,
+      "required_pages_per_second_upper_estimate": 0.002162316804722196
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28388.893574405538,
-      "required_pages_per_second_upper_estimate": 0.35249504667923487
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28088.322307996947,
+      "required_pages_per_second_upper_estimate": 0.34876295749775815
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2894173106877247,
-    "producer_p95_seconds": 0.4313449370019953,
-    "gpu_update_seconds": 0.9739019919017379,
-    "producer_batches_per_second": 3.4552183406851547,
-    "gpu_batches_per_second": 1.0267973659724225,
-    "producer_headroom": 3.36504402444869,
+    "producer_batch_seconds": 0.2562543832188453,
+    "producer_p95_seconds": 0.4387784240025212,
+    "gpu_update_seconds": 0.9843236522577362,
+    "producer_batches_per_second": 3.9023722733592585,
+    "gpu_batches_per_second": 1.0159260094038247,
+    "producer_headroom": 3.8411973285823104,
     "sustainable": true,
-    "buffer_seconds": 60.38192349790775,
+    "buffer_seconds": 60.04374278772191,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 538605.0892061491,
-    "upload_bytes_per_second": 5778877.514367516,
+    "generation_bytes_per_second": 528420.2551434791,
+    "upload_bytes_per_second": 5921203.960331778,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.57211405886035,
+    "estimated_full_shards_per_hour": 113.38668575981407,
     "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": 0.8978742559993407,
-      "bytes": 2461928,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 0.9092525360028958,
       "bytes": 6015320,
@@ -562,6 +556,12 @@
       "bytes": 7471926,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.4992664259989397,
+      "bytes": 16065938,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -417,3 +417,4 @@
 {"step": 156700, "loss": 0.6271420363336802, "lr": 0.0003, "grad_norm": 0.14207540156982212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22933594966, "tokens_per_second": 149792.27462428136, "data_wait_s": 1.5994002751540393e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156800, "loss": 0.61750128865242, "lr": 0.0003, "grad_norm": 0.15084095647069573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22948289405, "tokens_per_second": 149668.41289874725, "data_wait_s": 1.2961994798388332e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156900, "loss": 0.5993822142481804, "lr": 0.0003, "grad_norm": 0.14820837760259123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22963019955, "tokens_per_second": 150153.0592304569, "data_wait_s": 9.803996363189071e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 157000, "loss": 0.6234585996717215, "lr": 0.0003, "grad_norm": 0.15615076115668247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22977662403, "tokens_per_second": 148993.8608351398, "data_wait_s": 1.464000524720177e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c04a34888ea236f3d31688790d9cd40cb3e19a356f913077a94cf51de0b13286
-size 893871
+oid sha256:89e33e63c5ab46a03f24d4ff1ca552dc471a495dcc102568ff829194f82fcc47
+size 895989