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": 75253.47551861935,
-      "required_pages_per_second_upper_estimate": 0.006522780529763945
+      "required_token_positions_per_second": 76377.8677482956,
+      "required_pages_per_second_upper_estimate": 0.0066202399984862294
     },
     {
       "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": 9406.684439827419,
-      "required_pages_per_second_upper_estimate": 0.003752467065512773
+      "required_token_positions_per_second": 9547.23346853695,
+      "required_pages_per_second_upper_estimate": 0.003808534174460248
     },
     {
       "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": 18813.368879654838,
-      "required_pages_per_second_upper_estimate": 0.00435214178632447
+      "required_token_positions_per_second": 19094.4669370739,
+      "required_pages_per_second_upper_estimate": 0.0044171688747515255
     },
     {
       "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": 28220.053319482256,
-      "required_pages_per_second_upper_estimate": 0.0017184531879793183
+      "required_token_positions_per_second": 28641.70040561085,
+      "required_pages_per_second_upper_estimate": 0.0017441292833132573
     },
     {
       "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": 28220.053319482256,
-      "required_pages_per_second_upper_estimate": 0.0021724578226411137
+      "required_token_positions_per_second": 28641.70040561085,
+      "required_pages_per_second_upper_estimate": 0.0022049173825251307
     },
     {
       "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": 28220.053319482256,
-      "required_pages_per_second_upper_estimate": 0.3503986157850709
+      "required_token_positions_per_second": 28641.70040561085,
+      "required_pages_per_second_upper_estimate": 0.35563406143276816
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.15304470182030627,
-    "producer_p95_seconds": 0.3623442529933527,
-    "gpu_update_seconds": 0.9797288363347163,
-    "producer_batches_per_second": 6.534038670441044,
-    "gpu_batches_per_second": 1.0206905859187738,
-    "producer_headroom": 6.401586103157241,
+    "producer_batch_seconds": 0.39413603042157774,
+    "producer_p95_seconds": 0.3822081159960362,
+    "gpu_update_seconds": 0.9653058166401257,
+    "producer_batches_per_second": 2.5371950870119004,
+    "gpu_batches_per_second": 1.0359411315686795,
+    "producer_headroom": 2.4491691754433376,
     "sustainable": true,
-    "buffer_seconds": 60.74318785275241,
+    "buffer_seconds": 60.81426644832792,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 437749.4742694267,
-    "upload_bytes_per_second": 5221839.7089922745,
+    "generation_bytes_per_second": 402248.7431974242,
+    "upload_bytes_per_second": 5285435.157326066,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.93084689199544,
+    "estimated_full_shards_per_hour": 86.31321641866727,
     "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": 0.8690516380011104,
-      "bytes": 4841742,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.4456373529974371,
       "bytes": 7578448,
@@ -562,6 +556,12 @@
       "bytes": 11650354,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2925304770033108,
+      "bytes": 12240192,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -356,3 +356,4 @@
 {"step": 190600, "loss": 0.5875911097973585, "lr": 0.0003, "grad_norm": 0.13652470681259007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27896256213, "tokens_per_second": 150697.28122299162, "data_wait_s": 8.203998731914908e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 190700, "loss": 0.5992460511624813, "lr": 0.0003, "grad_norm": 0.14724762084252568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27910923120, "tokens_per_second": 151043.11295818255, "data_wait_s": 7.196998922154307e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 190800, "loss": 0.5666274484246969, "lr": 0.0003, "grad_norm": 0.13830076820219755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27925378730, "tokens_per_second": 148450.05855414632, "data_wait_s": 9.757997759152204e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 190900, "loss": 0.5981983579695225, "lr": 0.0003, "grad_norm": 0.149478788079187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27940072447, "tokens_per_second": 150759.55153794054, "data_wait_s": 9.71899862634018e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:329959e2c325c324bdf0ae940462276379760547891d7c02dbb02700a44bff26
-size 763890
+oid sha256:0de643aa42fc72752f4dc5623f3cd1f1f69521107bf5906bacc6fe5a7b78efcf
+size 766008