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": 74956.5513283916,
-      "required_pages_per_second_upper_estimate": 0.006497043893502483
+      "required_token_positions_per_second": 75425.27935402792,
+      "required_pages_per_second_upper_estimate": 0.006537672050784321
     },
     {
       "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": 9369.56891604895,
-      "required_pages_per_second_upper_estimate": 0.003737661128151009
+      "required_token_positions_per_second": 9428.15991925349,
+      "required_pages_per_second_upper_estimate": 0.0037610339553428634
     },
     {
       "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": 18739.1378320979,
-      "required_pages_per_second_upper_estimate": 0.004334969739893989
+      "required_token_positions_per_second": 18856.31983850698,
+      "required_pages_per_second_upper_estimate": 0.004362077734743855
     },
     {
       "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": 28108.706748146848,
-      "required_pages_per_second_upper_estimate": 0.001711672765975303
+      "required_token_positions_per_second": 28284.47975776047,
+      "required_pages_per_second_upper_estimate": 0.0017223764200510632
     },
     {
       "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": 28108.706748146848,
-      "required_pages_per_second_upper_estimate": 0.002163886055352677
+      "required_token_positions_per_second": 28284.47975776047,
+      "required_pages_per_second_upper_estimate": 0.0021774175482035664
     },
     {
       "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": 28108.706748146848,
-      "required_pages_per_second_upper_estimate": 0.34901606402208735
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28284.47975776047,
+      "required_pages_per_second_upper_estimate": 0.35119857652706793
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.0784813068828214,
-    "producer_p95_seconds": 0.3209245939979155,
-    "gpu_update_seconds": 0.9836098205344428,
-    "producer_batches_per_second": 12.741887714651293,
-    "gpu_batches_per_second": 1.0166632938421167,
-    "producer_headroom": 12.53304588827818,
+    "producer_batch_seconds": 0.4100162724452616,
+    "producer_p95_seconds": 0.33825653200256056,
+    "gpu_update_seconds": 0.9774972082494875,
+    "producer_batches_per_second": 2.438927591912838,
+    "gpu_batches_per_second": 1.0230208245717762,
+    "producer_headroom": 2.3840449122174445,
     "sustainable": true,
-    "buffer_seconds": 60.00019905260101,
+    "buffer_seconds": 60.60482691146822,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357778.4857156052,
-    "upload_bytes_per_second": 4003547.225777325,
+    "generation_bytes_per_second": 362732.6574649206,
+    "upload_bytes_per_second": 3964809.911052963,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.77093437768094,
+    "estimated_full_shards_per_hour": 77.83398430786814,
     "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": 1.0663465249999717,
-      "bytes": 5706618,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.3083136000022932,
       "bytes": 13235368,
@@ -562,6 +556,12 @@
       "bytes": 6207827,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9678468759993848,
+      "bytes": 1151038,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -240,3 +240,4 @@
 {"step": 139000, "loss": 0.589487548917532, "lr": 0.0003, "grad_norm": 0.1428047880376676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20342441712, "tokens_per_second": 150873.92026288502, "data_wait_s": 1.2471999070839956e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 139100, "loss": 0.574429651722312, "lr": 0.0003, "grad_norm": 0.1376758639669584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20357154219, "tokens_per_second": 150314.10776939144, "data_wait_s": 7.683000148972496e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 139200, "loss": 0.6169905085116625, "lr": 0.0003, "grad_norm": 0.12975897033810646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20371814263, "tokens_per_second": 149834.12120888062, "data_wait_s": 9.225001122104004e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 139300, "loss": 0.5964872129261494, "lr": 0.0003, "grad_norm": 0.1318150963052732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20386459708, "tokens_per_second": 149490.33986902668, "data_wait_s": 1.4132998330751434e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5058d1805d258789b824f83b3012c3e5b51fc39b2020c13a9fd14760200e234d
-size 515853
+oid sha256:5e202c83c388a94213f0b6ab519134ed9ffff4a597b2ef58f67a71d2de5088d5
+size 517971