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": 76013.07124090774,
-      "required_pages_per_second_upper_estimate": 0.006588620361794146
+      "required_token_positions_per_second": 75549.00253542721,
+      "required_pages_per_second_upper_estimate": 0.006548396062574474
     },
     {
       "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": 9501.633905113467,
-      "required_pages_per_second_upper_estimate": 0.003790343826836392
+      "required_token_positions_per_second": 9443.625316928401,
+      "required_pages_per_second_upper_estimate": 0.0037672033337036864
     },
     {
       "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": 19003.267810226935,
-      "required_pages_per_second_upper_estimate": 0.004396071561805294
+      "required_token_positions_per_second": 18887.250633856802,
+      "required_pages_per_second_upper_estimate": 0.0043692330298846305
     },
     {
       "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": 28504.9017153404,
-      "required_pages_per_second_upper_estimate": 0.0017357989608030458
+      "required_token_positions_per_second": 28330.875950785205,
+      "required_pages_per_second_upper_estimate": 0.0017252017047842574
     },
     {
       "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": 28504.9017153404,
-      "required_pages_per_second_upper_estimate": 0.0021943862406651003
+      "required_token_positions_per_second": 28330.875950785205,
+      "required_pages_per_second_upper_estimate": 0.0021809892555754926
     },
     {
       "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": 28504.9017153404,
-      "required_pages_per_second_upper_estimate": 0.3539354795353738
+      "required_token_positions_per_second": 28330.875950785205,
+      "required_pages_per_second_upper_estimate": 0.3517746619663658
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.10518032157042967,
-    "producer_p95_seconds": 0.31320514699837076,
-    "gpu_update_seconds": 0.9699384434334238,
-    "producer_batches_per_second": 9.507481866086435,
-    "gpu_batches_per_second": 1.0309932622735967,
-    "producer_headroom": 9.221672162163381,
+    "producer_batch_seconds": 0.09494392303909649,
+    "producer_p95_seconds": 0.31199331099924166,
+    "gpu_update_seconds": 0.9758964053221842,
+    "producer_batches_per_second": 10.532532973049944,
+    "gpu_batches_per_second": 1.024698927618099,
+    "producer_headroom": 10.278661067336818,
     "sustainable": true,
-    "buffer_seconds": 60.13618349287228,
+    "buffer_seconds": 60.50557712997542,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 469872.45421547524,
-    "upload_bytes_per_second": 5103516.557222069,
+    "generation_bytes_per_second": 477122.72213218,
+    "upload_bytes_per_second": 5118872.38862137,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.82369060371583,
+    "estimated_full_shards_per_hour": 102.37942932104158,
     "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": 1.2797600039993995,
-      "bytes": 6001788,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0512617659987882,
       "bytes": 6612201,
@@ -562,6 +556,12 @@
       "bytes": 12385340,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.0688240559975384,
+      "bytes": 6186249,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -295,3 +295,4 @@
 {"step": 384500, "loss": 0.6108637005090714, "lr": 0.0003, "grad_norm": 0.13859899117972965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56279962069, "tokens_per_second": 150048.15915098359, "data_wait_s": 8.611998055130243e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 384600, "loss": 0.5709079038351774, "lr": 0.0003, "grad_norm": 0.1481559462271218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56294668692, "tokens_per_second": 149785.23607901117, "data_wait_s": 9.123999916482717e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 384700, "loss": 0.6731205899268389, "lr": 0.0003, "grad_norm": 0.1632658477934949, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56309473763, "tokens_per_second": 150441.9841663838, "data_wait_s": 1.1893000191776082e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 384800, "loss": 0.6171527206897736, "lr": 0.0003, "grad_norm": 0.15566373443762968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56324230617, "tokens_per_second": 149848.2530638584, "data_wait_s": 1.133399928221479e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af57eaa87476e4fbbcad883cc108420fe5c7a43517e5ac4603e11bb43aa7f1c9
-size 633126
+oid sha256:54fab8424efe1ad6de5dd867d02c5bff28449b54f2f690fd6b21b946136efbfa
+size 635244