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": 75484.1813552648,
-      "required_pages_per_second_upper_estimate": 0.006542777526965775
+      "required_token_positions_per_second": 75762.12073546513,
+      "required_pages_per_second_upper_estimate": 0.006566868608010621
     },
     {
       "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": 9435.5226694081,
-      "required_pages_per_second_upper_estimate": 0.003763971066462462
+      "required_token_positions_per_second": 9470.26509193314,
+      "required_pages_per_second_upper_estimate": 0.003777830338253208
     },
     {
       "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": 18871.0453388162,
-      "required_pages_per_second_upper_estimate": 0.0043654842200804385
+      "required_token_positions_per_second": 18940.53018386628,
+      "required_pages_per_second_upper_estimate": 0.004381558316091269
     },
     {
       "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": 28306.568008224298,
-      "required_pages_per_second_upper_estimate": 0.0017237214786162166
+      "required_token_positions_per_second": 28410.79527579942,
+      "required_pages_per_second_upper_estimate": 0.0017300683723733305
     },
     {
       "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": 28306.568008224298,
-      "required_pages_per_second_upper_estimate": 0.0021791179628684624
+      "required_token_positions_per_second": 28410.79527579942,
+      "required_pages_per_second_upper_estimate": 0.002187141666445938
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28306.568008224298,
-      "required_pages_per_second_upper_estimate": 0.35147283867320983
+      "required_token_positions_per_second": 28410.79527579942,
+      "required_pages_per_second_upper_estimate": 0.3527669925102676
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05871126401557092,
-    "producer_p95_seconds": 0.20288309500028845,
-    "gpu_update_seconds": 0.976734445234302,
-    "producer_batches_per_second": 17.032506738992847,
-    "gpu_batches_per_second": 1.0238197340937607,
-    "producer_headroom": 16.63623602065969,
+    "producer_batch_seconds": 0.14025739783589586,
+    "producer_p95_seconds": 0.24670027799948002,
+    "gpu_update_seconds": 0.9731512170499086,
+    "producer_batches_per_second": 7.1297487008137805,
+    "gpu_batches_per_second": 1.0275895282045509,
+    "producer_headroom": 6.938323625456935,
     "sustainable": true,
-    "buffer_seconds": 60.55753560452672,
+    "buffer_seconds": 60.33537545709433,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 209282.20000964,
-    "upload_bytes_per_second": 6482928.945043092,
+    "generation_bytes_per_second": 220112.59193438836,
+    "upload_bytes_per_second": 6530977.281327566,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 44.90708828179264,
+    "estimated_full_shards_per_hour": 47.231038270223024,
     "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": 1.0638150449999557,
-      "bytes": 7164962,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.0547767480002221,
       "bytes": 2359076,
@@ -562,6 +556,12 @@
       "bytes": 8851756,
       "pages": 43,
       "shards": 4
+    },
+    {
+      "seconds": 1.0367009409983439,
+      "bytes": 10231090,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -106,3 +106,4 @@
 {"step": 10600, "loss": 0.7878618203103542, "lr": 0.0003, "grad_norm": 0.15678304360026926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1550485398, "tokens_per_second": 150077.2938703684, "data_wait_s": 1.0660998668754473e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 10700, "loss": 0.747017864137888, "lr": 0.0003, "grad_norm": 0.15221159743124865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1565188988, "tokens_per_second": 148680.92178058103, "data_wait_s": 1.1062000339734368e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 10800, "loss": 0.7142677288502455, "lr": 0.0003, "grad_norm": 0.16578387173052617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1579822809, "tokens_per_second": 149322.5279770065, "data_wait_s": 6.496000423794612e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 10900, "loss": 0.6950379349291325, "lr": 0.0003, "grad_norm": 0.15290236586700576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1594471529, "tokens_per_second": 148803.07798327558, "data_wait_s": 1.2014999811071903e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9668ce7a334d5a7bf0f1597fe482b931fc41aea6e7ab78487bb809081b9b1754
-size 226636
+oid sha256:15baef588cad4bdda45acc9f65fc041f1ae4feaf56cd5a09bcfdbffc877269f8
+size 228719