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": 78853.96613762408,
-      "required_pages_per_second_upper_estimate": 0.00683486193125923
+      "required_token_positions_per_second": 69597.10018760353,
+      "required_pages_per_second_upper_estimate": 0.0060325002520237045
     },
     {
       "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": 9856.74576720301,
-      "required_pages_per_second_upper_estimate": 0.003932003258019391
+      "required_token_positions_per_second": 8699.637523450441,
+      "required_pages_per_second_upper_estimate": 0.0034704154792765444
     },
     {
       "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": 19713.49153440602,
-      "required_pages_per_second_upper_estimate": 0.004560369320883491
+      "required_token_positions_per_second": 17399.275046900882,
+      "required_pages_per_second_upper_estimate": 0.004025016065318295
     },
     {
       "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": 29570.23730160903,
-      "required_pages_per_second_upper_estimate": 0.0018006723086229548
+      "required_token_positions_per_second": 26098.912570351327,
+      "required_pages_per_second_upper_estimate": 0.001589286845122679
     },
     {
       "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": 29570.23730160903,
-      "required_pages_per_second_upper_estimate": 0.0022763987231337094
+      "required_token_positions_per_second": 26098.912570351327,
+      "required_pages_per_second_upper_estimate": 0.002009166536079616
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 29570.23730160903,
-      "required_pages_per_second_upper_estimate": 0.36716338206798155
+      "estimated_pages_for_buffer": 20,
+      "required_token_positions_per_second": 26098.912570351327,
+      "required_pages_per_second_upper_estimate": 0.32406114668228675
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 65,
+    "prefetch_batches": 57,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 2.3582960475177965,
-    "producer_p95_seconds": 17.862166522000052,
-    "gpu_update_seconds": 0.9349941874999956,
-    "producer_batches_per_second": 0.4240349726458394,
-    "gpu_batches_per_second": 1.06952536536491,
-    "producer_headroom": 0.39647023472057946,
-    "sustainable": false,
-    "buffer_seconds": 60.774622187499716,
-    "buffer_drain_seconds": 100.69863274989008,
+    "producer_batch_seconds": 0.04300748954687528,
+    "producer_p95_seconds": 0.1263296930000024,
+    "gpu_update_seconds": 1.0593544817422185,
+    "producer_batches_per_second": 23.25176406565343,
+    "gpu_batches_per_second": 0.9439710854438413,
+    "producer_headroom": 24.631860471362625,
+    "sustainable": true,
+    "buffer_seconds": 60.38320545930645,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 15,
-    "max_pending_pages": 1842,
-    "generation_bytes_per_second": 1035119.2626246791,
-    "upload_bytes_per_second": 6146691.738378885,
+    "upload_pages_per_commit": 82,
+    "max_pending_pages": 4096,
+    "generation_bytes_per_second": 910173.0309263988,
+    "upload_bytes_per_second": 3293326.3783964273,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.0562487080349,
+    "estimated_full_shards_per_hour": 97.65097234651553,
     "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": [
@@ -184,6 +184,30 @@
       "bytes": 5047481,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 2.1621551530000716,
+      "bytes": 13914829,
+      "pages": 15,
+      "shards": 5
+    },
+    {
+      "seconds": 0.7754683500000965,
+      "bytes": 204111,
+      "pages": 6,
+      "shards": 1
+    },
+    {
+      "seconds": 1.1240084219999744,
+      "bytes": 42370,
+      "pages": 1,
+      "shards": 1
+    },
+    {
+      "seconds": 1.1000538829999869,
+      "bytes": 494707,
+      "pages": 15,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -0,0 +1 @@
+{"step": 100, "loss": 2.8035809621214867, "lr": 2.9999999999999997e-05, "grad_norm": 1.586405728010338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14758037, "tokens_per_second": 46373.26368537362, "data_wait_s": 8.257999979832675e-06, "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:a10197bec45f649a59459d8677528d5ac83c3578cace0d8df7d92c9d62088448
-size 161
+oid sha256:4430eb84436ec72df1c6721c0d95aac3150cb06c67c25b1cf73c5b969e6be77c
+size 2209