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": 73231.98518460449,
-      "required_pages_per_second_upper_estimate": 0.006347562871031948
+      "required_token_positions_per_second": 73533.88586639984,
+      "required_pages_per_second_upper_estimate": 0.006373730856969686
     },
     {
       "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": 9153.998148075561,
-      "required_pages_per_second_upper_estimate": 0.003651666725736222
+      "required_token_positions_per_second": 9191.73573329998,
+      "required_pages_per_second_upper_estimate": 0.0036667208127094224
     },
     {
       "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": 18307.996296151123,
-      "required_pages_per_second_upper_estimate": 0.004235232733384579
+      "required_token_positions_per_second": 18383.47146659996,
+      "required_pages_per_second_upper_estimate": 0.004252692585750285
     },
     {
       "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": 27461.994444226682,
-      "required_pages_per_second_upper_estimate": 0.001672291379703796
+      "required_token_positions_per_second": 27575.207199899938,
+      "required_pages_per_second_upper_estimate": 0.0016791854425428734
     },
     {
       "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": 27461.994444226682,
-      "required_pages_per_second_upper_estimate": 0.002114100351982667
+      "required_token_positions_per_second": 27575.207199899938,
+      "required_pages_per_second_upper_estimate": 0.0021228157833073596
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27461.994444226682,
-      "required_pages_per_second_upper_estimate": 0.3409860616142479
+      "required_token_positions_per_second": 27575.207199899938,
+      "required_pages_per_second_upper_estimate": 0.34239178514099033
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.14769628186007822,
-    "producer_p95_seconds": 0.3928780080022989,
-    "gpu_update_seconds": 1.006773199089785,
-    "producer_batches_per_second": 6.7706511457570855,
-    "gpu_batches_per_second": 0.9932723684977822,
-    "producer_headroom": 6.816510113934779,
+    "producer_batch_seconds": 0.136012479656074,
+    "producer_p95_seconds": 0.4003450110030826,
+    "gpu_update_seconds": 1.0026397915914962,
+    "producer_batches_per_second": 7.352266516489044,
+    "gpu_batches_per_second": 0.9973671585611957,
+    "producer_headroom": 7.37167496781771,
     "sustainable": true,
-    "buffer_seconds": 60.4063919453871,
+    "buffer_seconds": 60.15838749548977,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 461466.9836228392,
-    "upload_bytes_per_second": 3785887.0474941544,
+    "generation_bytes_per_second": 461148.4747449724,
+    "upload_bytes_per_second": 3775662.4136489676,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.0200722838772,
+    "estimated_full_shards_per_hour": 98.9517276931942,
     "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": 1.0333191259996966,
-      "bytes": 7473380,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8470484819990816,
       "bytes": 1045811,
@@ -562,6 +556,12 @@
       "bytes": 4661656,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.2476324340022984,
+      "bytes": 7151046,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -366,3 +366,4 @@
 {"step": 351600, "loss": 0.5729933995753527, "lr": 0.0003, "grad_norm": 0.1799935482318668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51465136218, "tokens_per_second": 147513.37203497015, "data_wait_s": 1.3928001862950623e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 351700, "loss": 0.6278344821184874, "lr": 0.0003, "grad_norm": 0.1583050349370463, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51479621936, "tokens_per_second": 146789.8006376949, "data_wait_s": 1.997900108108297e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 351800, "loss": 0.5804985966533422, "lr": 0.0003, "grad_norm": 0.17546815883735956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51494197641, "tokens_per_second": 148130.76413141593, "data_wait_s": 1.1023999832104892e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 351900, "loss": 0.5837832074612379, "lr": 0.0003, "grad_norm": 0.15345544160138913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51508829587, "tokens_per_second": 148419.20414938993, "data_wait_s": 1.3760000001639128e-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:ef65a6465e2dea24c2a851842ec2479df9c6352689bfac8b8bea5e1642de8a17
-size 785070
+oid sha256:c19709c44ede09e27859dc2119f0c9babb3783bc26fbd941e1c16412f6482e5b
+size 787188