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": 75801.08895547003,
-      "required_pages_per_second_upper_estimate": 0.006570246274556595
+      "required_token_positions_per_second": 75601.84365788664,
+      "required_pages_per_second_upper_estimate": 0.006552976197144657
     },
     {
       "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": 9475.136119433753,
-      "required_pages_per_second_upper_estimate": 0.003779773463951553
+      "required_token_positions_per_second": 9450.23045723583,
+      "required_pages_per_second_upper_estimate": 0.003769838222927968
     },
     {
       "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": 18950.272238867507,
-      "required_pages_per_second_upper_estimate": 0.00438381196906149
+      "required_token_positions_per_second": 18900.46091447166,
+      "required_pages_per_second_upper_estimate": 0.004372288995811879
     },
     {
       "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": 28425.408358301258,
-      "required_pages_per_second_upper_estimate": 0.0017309582324287738
+      "required_token_positions_per_second": 28350.69137170749,
+      "required_pages_per_second_upper_estimate": 0.0017264083599549459
     },
     {
       "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": 28425.408358301258,
-      "required_pages_per_second_upper_estimate": 0.002188266621988531
+      "required_token_positions_per_second": 28350.69137170749,
+      "required_pages_per_second_upper_estimate": 0.002182514701530686
     },
     {
       "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": 28425.408358301258,
-      "required_pages_per_second_upper_estimate": 0.35294843808809934
+      "required_token_positions_per_second": 28350.69137170749,
+      "required_pages_per_second_upper_estimate": 0.3520207031762729
     }
   ],
   "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": 0.12019333556253287,
-    "producer_p95_seconds": 0.3807493739950587,
-    "gpu_update_seconds": 0.9726509343858124,
-    "producer_batches_per_second": 8.3199288489646,
-    "gpu_batches_per_second": 1.0281180685149471,
-    "producer_headroom": 8.092386568968895,
+    "producer_batch_seconds": 0.12646835916433474,
+    "producer_p95_seconds": 0.38606119099858915,
+    "gpu_update_seconds": 0.9752143126777946,
+    "producer_batches_per_second": 7.907116108785646,
+    "gpu_batches_per_second": 1.025415631210485,
+    "producer_headroom": 7.711132801292911,
     "sustainable": true,
-    "buffer_seconds": 60.30435793192037,
+    "buffer_seconds": 60.46328738602326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374297.89681714645,
-    "upload_bytes_per_second": 6026828.365055565,
+    "generation_bytes_per_second": 375010.7815569023,
+    "upload_bytes_per_second": 6030665.662634119,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.31561544786258,
+    "estimated_full_shards_per_hour": 80.46858391790677,
     "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": 0.8729745059972629,
-      "bytes": 5859266,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0330235869987519,
       "bytes": 6506267,
@@ -562,6 +556,12 @@
       "bytes": 1141313,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 0.8455009660028736,
+      "bytes": 5959516,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -370,3 +370,4 @@
 {"step": 77000, "loss": 0.6380148641765118, "lr": 0.0003, "grad_norm": 0.14502346248895828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11268718373, "tokens_per_second": 150871.65479264862, "data_wait_s": 8.916998922359198e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 77100, "loss": 0.6637343280017376, "lr": 0.0003, "grad_norm": 0.13113832431376676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11283422338, "tokens_per_second": 149391.6201590513, "data_wait_s": 1.138600055128336e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 77200, "loss": 0.6842169724404812, "lr": 0.0003, "grad_norm": 0.1527218037417005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11298186788, "tokens_per_second": 150424.5119888574, "data_wait_s": 1.251600042451173e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 77300, "loss": 0.6141677498817444, "lr": 0.0003, "grad_norm": 0.13769632136882737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11312831331, "tokens_per_second": 147915.06908929613, "data_wait_s": 8.301998605020344e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d3e3792d5a73e232d3b38f9d6e7a2d1a7db96a3b418e91b422ea4ee8f5695afb
-size 793542
+oid sha256:08e73065cfa5490f3130d04b1cb0d001a66ffb3f08cd227e6e6fca7a1e57a079
+size 795660