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": 75708.07424569054,
-      "required_pages_per_second_upper_estimate": 0.006562183995256536
+      "required_token_positions_per_second": 73883.53805485177,
+      "required_pages_per_second_upper_estimate": 0.00640403782247933
     },
     {
       "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": 9463.509280711318,
-      "required_pages_per_second_upper_estimate": 0.0037751353441484435
+      "required_token_positions_per_second": 9235.442256856471,
+      "required_pages_per_second_upper_estimate": 0.0036841559984268674
     },
     {
       "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": 18927.018561422636,
-      "required_pages_per_second_upper_estimate": 0.004378432640035365
+      "required_token_positions_per_second": 18470.884513712943,
+      "required_pages_per_second_upper_estimate": 0.004272914055782788
     },
     {
       "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": 28390.527842133954,
-      "required_pages_per_second_upper_estimate": 0.00172883419199811
+      "required_token_positions_per_second": 27706.326770569412,
+      "required_pages_per_second_upper_estimate": 0.0016871699364654233
     },
     {
       "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": 28390.527842133954,
-      "required_pages_per_second_upper_estimate": 0.002185581423298523
+      "required_token_positions_per_second": 27706.326770569412,
+      "required_pages_per_second_upper_estimate": 0.0021329097308196983
     },
     {
       "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": 28390.527842133954,
-      "required_pages_per_second_upper_estimate": 0.35251533881487956
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27706.326770569412,
+      "required_pages_per_second_upper_estimate": 0.34401985137973123
     }
   ],
   "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": 0.525914920163757,
-    "producer_p95_seconds": 0.37546664899855386,
-    "gpu_update_seconds": 0.9738459303658321,
-    "producer_batches_per_second": 1.9014482412642422,
-    "gpu_batches_per_second": 1.0268564757716274,
-    "producer_headroom": 1.8517176315564512,
+    "producer_batch_seconds": 0.12980772744543856,
+    "producer_p95_seconds": 0.3699612250056816,
+    "gpu_update_seconds": 0.9978948212423681,
+    "producer_batches_per_second": 7.703701618382658,
+    "gpu_batches_per_second": 1.0021096198845998,
+    "producer_headroom": 7.687483949380504,
     "sustainable": true,
-    "buffer_seconds": 60.37844768268159,
+    "buffer_seconds": 60.87158409578445,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301146.7535108694,
-    "upload_bytes_per_second": 5097266.34712741,
+    "generation_bytes_per_second": 302283.3007663712,
+    "upload_bytes_per_second": 5117133.181698163,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.61908296579897,
+    "estimated_full_shards_per_hour": 64.86295954936364,
     "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": 1.2969893279987446,
-      "bytes": 7578437,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.5002685959989321,
       "bytes": 6193769,
@@ -562,6 +556,12 @@
       "bytes": 1097321,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.6297084739999264,
+      "bytes": 10885111,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -394,3 +394,4 @@
 {"step": 276500, "loss": 0.6283031068742275, "lr": 0.0003, "grad_norm": 0.15542361558506881, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40471107394, "tokens_per_second": 147453.32761461716, "data_wait_s": 1.09350003185682e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276600, "loss": 0.580127177760005, "lr": 0.0003, "grad_norm": 0.14188547742678123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40485662186, "tokens_per_second": 142526.23012206602, "data_wait_s": 1.4037002983968705e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276700, "loss": 0.6000665910542011, "lr": 0.0003, "grad_norm": 0.1562039063917249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40500433646, "tokens_per_second": 150322.58836622757, "data_wait_s": 1.383500057272613e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 276800, "loss": 0.623013187199831, "lr": 0.0003, "grad_norm": 0.17697419124575287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40515274660, "tokens_per_second": 151072.97806939497, "data_wait_s": 1.374099520035088e-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:ee0b2e9e4b43391cd22665b7ba917ae5bba6d037617813c1540f9b3ad8b621ae
-size 845157
+oid sha256:63ec0869882caf6b28892d40b2faf7a1b308045bfa39515ad9356ba716cc7f23
+size 847275