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": 75655.3096847071,
-      "required_pages_per_second_upper_estimate": 0.0065576104968410485
+      "required_token_positions_per_second": 74645.64571712035,
+      "required_pages_per_second_upper_estimate": 0.0064700953831005594
     },
     {
       "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": 9456.913710588387,
-      "required_pages_per_second_upper_estimate": 0.0037725042726138454
+      "required_token_positions_per_second": 9330.705714640044,
+      "required_pages_per_second_upper_estimate": 0.0037221580160523553
     },
     {
       "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": 18913.827421176775,
-      "required_pages_per_second_upper_estimate": 0.004375381101895624
+      "required_token_positions_per_second": 18661.411429280088,
+      "required_pages_per_second_upper_estimate": 0.004316989104540057
     },
     {
       "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": 28370.741131765164,
-      "required_pages_per_second_upper_estimate": 0.001727629285149497
+      "required_token_positions_per_second": 27992.11714392013,
+      "required_pages_per_second_upper_estimate": 0.0017045730707762748
     },
     {
       "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": 28370.741131765164,
-      "required_pages_per_second_upper_estimate": 0.002184058187561214
+      "required_token_positions_per_second": 27992.11714392013,
+      "required_pages_per_second_upper_estimate": 0.002154910664878624
     },
     {
       "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": 28370.741131765164,
-      "required_pages_per_second_upper_estimate": 0.35226965409395883
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27992.11714392013,
+      "required_pages_per_second_upper_estimate": 0.347568411337896
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.12121701978912824,
-    "producer_p95_seconds": 0.28864622699984466,
-    "gpu_update_seconds": 0.9745251233160085,
-    "producer_batches_per_second": 8.249666604076076,
-    "gpu_batches_per_second": 1.0261408106107193,
-    "producer_headroom": 8.039507364653193,
+    "producer_batch_seconds": 0.06511146840628612,
+    "producer_p95_seconds": 0.21008682499814313,
+    "gpu_update_seconds": 0.9877066410464463,
+    "producer_batches_per_second": 15.358277496679156,
+    "gpu_batches_per_second": 1.0124463666059076,
+    "producer_headroom": 15.169472678504194,
     "sustainable": true,
-    "buffer_seconds": 60.420557645592524,
+    "buffer_seconds": 60.25010510383323,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353031.99985658133,
-    "upload_bytes_per_second": 5825030.05740388,
+    "generation_bytes_per_second": 352371.6109790492,
+    "upload_bytes_per_second": 5877157.604565401,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.75244900487023,
+    "estimated_full_shards_per_hour": 75.6107449248181,
     "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": 0.9418964010001218,
-      "bytes": 1018742,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.3071213839994016,
       "bytes": 12190710,
@@ -562,6 +556,12 @@
       "bytes": 7348514,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1231975430018792,
+      "bytes": 5800198,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -179,3 +179,4 @@
 {"step": 332900, "loss": 0.6819286067038774, "lr": 0.0003, "grad_norm": 0.2294065775638761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48728128703, "tokens_per_second": 146969.44527870012, "data_wait_s": 1.3622000551549718e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 333000, "loss": 0.5949093159288168, "lr": 0.0003, "grad_norm": 0.1557018599205989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48742868155, "tokens_per_second": 150064.95297595364, "data_wait_s": 1.3176999345887452e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 333100, "loss": 0.5879385620355606, "lr": 0.0003, "grad_norm": 0.15587732033841645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48757615375, "tokens_per_second": 150084.8878891177, "data_wait_s": 1.2298001820454374e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 333200, "loss": 0.570095119997859, "lr": 0.0003, "grad_norm": 0.16794487178184855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48772284324, "tokens_per_second": 148634.77690898417, "data_wait_s": 1.9146998965879902e-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:c693c24b4c6264ef4561ad59d1f5fce885c386639f20cb7b10223783397654c8
-size 385872
+oid sha256:4ebdd5a184df7b9beb820ac103c4652d5ec4a9a90d73b2f75a8f1a2e824952bb
+size 387990