Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77040.84845985073,
-      "required_pages_per_second_upper_estimate": 0.005329864655849139
+      "required_token_positions_per_second": 76139.39626286365,
+      "required_pages_per_second_upper_estimate": 0.0052675001012043015
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9630.106057481342,
-      "required_pages_per_second_upper_estimate": 0.004031924306785173
+      "required_token_positions_per_second": 9517.424532857956,
+      "required_pages_per_second_upper_estimate": 0.003984746905483171
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19260.212114962684,
-      "required_pages_per_second_upper_estimate": 0.0040113942575967145
+      "required_token_positions_per_second": 19034.849065715913,
+      "required_pages_per_second_upper_estimate": 0.003964457077661886
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28890.318172444026,
-      "required_pages_per_second_upper_estimate": 0.0017319304614763934
+      "required_token_positions_per_second": 28552.273598573873,
+      "required_pages_per_second_upper_estimate": 0.0017116652054370552
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28890.318172444026,
-      "required_pages_per_second_upper_estimate": 0.002246095162827591
+      "required_token_positions_per_second": 28552.273598573873,
+      "required_pages_per_second_upper_estimate": 0.0022198136841101264
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28890.318172444026,
-      "required_pages_per_second_upper_estimate": 0.3705739815092678
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28552.273598573873,
+      "required_pages_per_second_upper_estimate": 0.3662379086796459
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08286802085129352,
-    "producer_p95_seconds": 0.248200985002768,
-    "gpu_update_seconds": 0.9569988061388343,
-    "producer_batches_per_second": 12.06738123738345,
-    "gpu_batches_per_second": 1.044933382973236,
-    "producer_headroom": 11.548469437398131,
+    "producer_batch_seconds": 0.12514990897639677,
+    "producer_p95_seconds": 0.24740169199867523,
+    "gpu_update_seconds": 0.968329191178,
+    "producer_batches_per_second": 7.9904173177513025,
+    "gpu_batches_per_second": 1.032706655041011,
+    "producer_headroom": 7.737354338472803,
     "sustainable": true,
-    "buffer_seconds": 60.29092478674656,
+    "buffer_seconds": 60.036409853036,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 218649.25064497153,
-    "upload_bytes_per_second": 5125443.499213305,
+    "generation_bytes_per_second": 245006.63000865898,
+    "upload_bytes_per_second": 5074193.710142839,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 46.91703929435596,
+    "estimated_full_shards_per_hour": 52.57271933741404,
     "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": 1.290804393000144,
-      "bytes": 6092692,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4008768420007982,
       "bytes": 11876531,
@@ -562,6 +556,12 @@
       "bytes": 10704029,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.136130429997138,
+      "bytes": 965107,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -189,3 +189,4 @@
 {"step": 903900, "loss": 0.5374611020088196, "lr": 3e-05, "grad_norm": 0.33639597405785376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132304702741, "tokens_per_second": 151373.51801883074, "data_wait_s": 1.7398000636603683e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 904000, "loss": 0.5344279408454895, "lr": 3e-05, "grad_norm": 0.32576334446273497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132319457810, "tokens_per_second": 154039.98092213096, "data_wait_s": 1.3061999197816476e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 904100, "loss": 0.529141902923584, "lr": 3e-05, "grad_norm": 0.3101500490813347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132334192355, "tokens_per_second": 153180.37902358215, "data_wait_s": 1.322599928244017e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 904200, "loss": 0.5523757934570312, "lr": 3e-05, "grad_norm": 0.32844227088961897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132348936461, "tokens_per_second": 153840.23870132203, "data_wait_s": 1.5313999028876424e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:497765616ba89ffa4a7a833895514386505ca681e52830ceb0a90b08ea6d4bc1
-size 407052
+oid sha256:acf17eb5e6ea30e0148424fa6edceff030397d8963ed0d70f1600554f18bb08f
+size 409170