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": 75695.80589391947,
-      "required_pages_per_second_upper_estimate": 0.006561120605618873
+      "required_token_positions_per_second": 75147.90417618312,
+      "required_pages_per_second_upper_estimate": 0.006513629873369693
     },
     {
       "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": 9461.975736739934,
-      "required_pages_per_second_upper_estimate": 0.003774523590529733
+      "required_token_positions_per_second": 9393.48802202289,
+      "required_pages_per_second_upper_estimate": 0.0037472028171465174
     },
     {
       "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": 18923.95147347987,
-      "required_pages_per_second_upper_estimate": 0.0043777231232187104
+      "required_token_positions_per_second": 18786.97604404578,
+      "required_pages_per_second_upper_estimate": 0.004346036268304356
     },
     {
       "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": 28385.9272102198,
-      "required_pages_per_second_upper_estimate": 0.0017285540376521882
+      "required_token_positions_per_second": 28180.46406606867,
+      "required_pages_per_second_upper_estimate": 0.001716042410155191
     },
     {
       "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": 28385.9272102198,
-      "required_pages_per_second_upper_estimate": 0.0021852272539184077
+      "required_token_positions_per_second": 28180.46406606867,
+      "required_pages_per_second_upper_estimate": 0.0021694101323232683
     },
     {
       "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": 28385.9272102198,
-      "required_pages_per_second_upper_estimate": 0.3524582143638303
+      "required_token_positions_per_second": 28180.46406606867,
+      "required_pages_per_second_upper_estimate": 0.3499070497543821
     }
   ],
   "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": 0.07384667560948799,
-    "producer_p95_seconds": 0.18275351199918077,
-    "gpu_update_seconds": 0.9740037658535908,
-    "producer_batches_per_second": 13.54157098808545,
-    "gpu_batches_per_second": 1.0266900756011212,
-    "producer_headroom": 13.189541137968959,
+    "producer_batch_seconds": 0.868930449781331,
+    "producer_p95_seconds": 0.1885331329995097,
+    "gpu_update_seconds": 0.9811052059036247,
+    "producer_batches_per_second": 1.1508400934178944,
+    "gpu_batches_per_second": 1.0192586829451922,
+    "producer_headroom": 1.12909520681491,
     "sustainable": true,
-    "buffer_seconds": 60.38823348292263,
+    "buffer_seconds": 60.82852276602473,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 486354.953512172,
-    "upload_bytes_per_second": 3992148.6225429447,
+    "generation_bytes_per_second": 324616.17385700915,
+    "upload_bytes_per_second": 3973592.222472925,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.36045125984067,
+    "estimated_full_shards_per_hour": 69.65507423193651,
     "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": 1.2959069240000645,
-      "bytes": 8294236,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.3743943120000495,
       "bytes": 10663078,
@@ -562,6 +556,12 @@
       "bytes": 1136030,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.1651587139986077,
+      "bytes": 5801842,
+      "pages": 20,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -87,3 +87,4 @@
 {"step": 323700, "loss": 0.6073876842856407, "lr": 0.0003, "grad_norm": 0.1726146494891655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47380602911, "tokens_per_second": 150101.78471536125, "data_wait_s": 1.389600038237404e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 323800, "loss": 0.6049603596329689, "lr": 0.0003, "grad_norm": 0.14880438064026513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47395190148, "tokens_per_second": 148905.275910344, "data_wait_s": 1.760400118655525e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 323900, "loss": 0.5672470610588789, "lr": 0.0003, "grad_norm": 0.15293876371581858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47409874860, "tokens_per_second": 89906.31058175236, "data_wait_s": 1.6027999663492665e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 324000, "loss": 0.6469105780124664, "lr": 0.0003, "grad_norm": 0.17834168760019067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47424679180, "tokens_per_second": 150837.03804419088, "data_wait_s": 1.5948000509524718e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8962e681f5dbb184919eb2b97dc95c99d78fbb1594d000d3931183e5201eb1f1
-size 189450
+oid sha256:68394be45f9718dba2856a4be22bd1bbbfb17886c5b343a18fa5b4260effc57d
+size 191568