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": 74511.88958778027,
-      "required_pages_per_second_upper_estimate": 0.006458501740811172
+      "required_token_positions_per_second": 75367.02494118852,
+      "required_pages_per_second_upper_estimate": 0.00653262270592386
     },
     {
       "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": 9313.986198472534,
-      "required_pages_per_second_upper_estimate": 0.003715488351074092
+      "required_token_positions_per_second": 9420.878117648565,
+      "required_pages_per_second_upper_estimate": 0.0037581291358100228
     },
     {
       "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": 18627.972396945068,
-      "required_pages_per_second_upper_estimate": 0.0043092535729162195
+      "required_token_positions_per_second": 18841.75623529713,
+      "required_pages_per_second_upper_estimate": 0.0043587087014519185
     },
     {
       "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": 27941.9585954176,
-      "required_pages_per_second_upper_estimate": 0.0017015186783340365
+      "required_token_positions_per_second": 28262.634352945694,
+      "required_pages_per_second_upper_estimate": 0.0017210461495118247
     },
     {
       "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": 27941.9585954176,
-      "required_pages_per_second_upper_estimate": 0.00215104932096715
+      "required_token_positions_per_second": 28262.634352945694,
+      "required_pages_per_second_upper_estimate": 0.002175735828468974
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27941.9585954176,
-      "required_pages_per_second_upper_estimate": 0.3469456100353577
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28262.634352945694,
+      "required_pages_per_second_upper_estimate": 0.35092732971113516
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10623873908591008,
-    "producer_p95_seconds": 0.3320843760011485,
-    "gpu_update_seconds": 0.9894796710683764,
-    "producer_batches_per_second": 9.41276231819119,
-    "gpu_batches_per_second": 1.0106321829939815,
-    "producer_headroom": 9.313736962448628,
+    "producer_batch_seconds": 0.08014401413271344,
+    "producer_p95_seconds": 0.2940763770020567,
+    "gpu_update_seconds": 0.978252757854413,
+    "producer_batches_per_second": 12.477538226923137,
+    "gpu_batches_per_second": 1.0222306985295753,
+    "producer_headroom": 12.20618618172142,
     "sustainable": true,
-    "buffer_seconds": 60.35825993517096,
+    "buffer_seconds": 60.651670986973606,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291518.8804254749,
-    "upload_bytes_per_second": 4557782.441977322,
+    "generation_bytes_per_second": 299954.5702487405,
+    "upload_bytes_per_second": 4503026.7145899655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.55316552708803,
+    "estimated_full_shards_per_hour": 64.36326819035207,
     "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": 1.3090287019986135,
-      "bytes": 11650253,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.360405893999996,
       "bytes": 12362632,
@@ -562,6 +556,12 @@
       "bytes": 2407311,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.2172690299994429,
+      "bytes": 6177200,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -256,3 +256,4 @@
 {"step": 262700, "loss": 0.6104465872049332, "lr": 0.0003, "grad_norm": 0.1527788959809341, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38451981381, "tokens_per_second": 117888.91027919229, "data_wait_s": 1.3957000192021951e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 262800, "loss": 0.5883192215114832, "lr": 0.0003, "grad_norm": 0.13911527438268312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38466570218, "tokens_per_second": 148257.85721483117, "data_wait_s": 1.0211999324383214e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 262900, "loss": 0.596819169819355, "lr": 0.0003, "grad_norm": 0.15220561667072507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38481105543, "tokens_per_second": 147146.77668089542, "data_wait_s": 1.8528000509832054e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 263000, "loss": 0.6872336938977242, "lr": 0.0003, "grad_norm": 0.17677581162894507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38495777649, "tokens_per_second": 149705.37010786976, "data_wait_s": 1.454499943065457e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4215c146226e2cfbf753de6f43062fc2c1ea5134b0abd689d0e964c208d87b20
-size 550524
+oid sha256:1cfaaf76820d4c034007bc2e8836b774a0356acf9bad160889fc92dc02d11f00
+size 552642