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": 75817.29059799135,
-      "required_pages_per_second_upper_estimate": 0.006571650591867656
+      "required_token_positions_per_second": 74618.37475451561,
+      "required_pages_per_second_upper_estimate": 0.0064677316051796035
     },
     {
       "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": 9477.161324748919,
-      "required_pages_per_second_upper_estimate": 0.0037805813486312904
+      "required_token_positions_per_second": 9327.296844314451,
+      "required_pages_per_second_upper_estimate": 0.0037207981667123228
     },
     {
       "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": 18954.322649497837,
-      "required_pages_per_second_upper_estimate": 0.004384748960276025
+      "required_token_positions_per_second": 18654.593688628902,
+      "required_pages_per_second_upper_estimate": 0.004315411940228543
     },
     {
       "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": 28431.483974246756,
-      "required_pages_per_second_upper_estimate": 0.0017313282055635613
+      "required_token_positions_per_second": 27981.89053294335,
+      "required_pages_per_second_upper_estimate": 0.00170395032382267
     },
     {
       "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": 28431.483974246756,
-      "required_pages_per_second_upper_estimate": 0.002188734339722397
+      "required_token_positions_per_second": 27981.89053294335,
+      "required_pages_per_second_upper_estimate": 0.00215412339205656
     },
     {
       "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": 28431.483974246756,
-      "required_pages_per_second_upper_estimate": 0.3530238769043639
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27981.89053294335,
+      "required_pages_per_second_upper_estimate": 0.34744143105582964
     }
   ],
   "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": 0.05493099188284134,
-    "producer_p95_seconds": 0.1969234910011437,
-    "gpu_update_seconds": 0.9724430854556717,
-    "producer_batches_per_second": 18.204659441301068,
-    "gpu_batches_per_second": 1.0283378173555684,
-    "producer_headroom": 17.702995196768537,
+    "producer_batch_seconds": 0.4203233830468065,
+    "producer_p95_seconds": 0.21264051100115466,
+    "gpu_update_seconds": 0.9880676206437782,
+    "producer_batches_per_second": 2.379120554158277,
+    "gpu_batches_per_second": 1.0120764805028701,
+    "producer_headroom": 2.3507319851718758,
     "sustainable": true,
-    "buffer_seconds": 60.29147129825164,
+    "buffer_seconds": 60.27212485927047,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 281325.6006963566,
-    "upload_bytes_per_second": 3995746.374638745,
+    "generation_bytes_per_second": 264063.43674184015,
+    "upload_bytes_per_second": 4042786.830392722,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.365924984627,
+    "estimated_full_shards_per_hour": 56.661866442598374,
     "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": 1.3044787690000703,
-      "bytes": 7311631,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 0.7353289520001454,
       "bytes": 1217005,
@@ -562,6 +556,12 @@
       "bytes": 1018742,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.3071213839994016,
+      "bytes": 12190710,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -103,3 +103,4 @@
 {"step": 325300, "loss": 0.5691273473203182, "lr": 0.0003, "grad_norm": 0.1640405280227942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47614986024, "tokens_per_second": 147575.77974620298, "data_wait_s": 1.2975000572623685e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325400, "loss": 0.556862710043788, "lr": 0.0003, "grad_norm": 0.15268652971653798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47629441839, "tokens_per_second": 147466.7865420081, "data_wait_s": 1.2386999515001662e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325500, "loss": 0.6061256788671017, "lr": 0.0003, "grad_norm": 0.1477813010888525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47644090091, "tokens_per_second": 145756.85641014134, "data_wait_s": 1.6502001017215662e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 325600, "loss": 0.6151839569211006, "lr": 0.0003, "grad_norm": 0.15870963723315856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47658801353, "tokens_per_second": 149701.9202591019, "data_wait_s": 1.384799907100387e-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:d2f7c1dccc47d305d575a4af707f3343720e8c7ccfe5c27b9c9e74e95bdec46d
-size 224121
+oid sha256:deda554a7f51c0156d254fca2c49e95700ca72e115bb5d04c4f5321d18a7803a
+size 226239