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": 75739.99025522982,
-      "required_pages_per_second_upper_estimate": 0.006564950394073014
+      "required_token_positions_per_second": 75568.0522798349,
+      "required_pages_per_second_upper_estimate": 0.006550047246138596
     },
     {
       "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": 9467.498781903727,
-      "required_pages_per_second_upper_estimate": 0.0037767268158224536
+      "required_token_positions_per_second": 9446.006534979362,
+      "required_pages_per_second_upper_estimate": 0.0037681532371865974
     },
     {
       "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": 18934.997563807454,
-      "required_pages_per_second_upper_estimate": 0.004380278441811447
+      "required_token_positions_per_second": 18892.013069958724,
+      "required_pages_per_second_upper_estimate": 0.004370334735660797
     },
     {
       "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": 28402.49634571118,
-      "required_pages_per_second_upper_estimate": 0.0017295630110720777
+      "required_token_positions_per_second": 28338.01960493809,
+      "required_pages_per_second_upper_estimate": 0.0017256367158423114
     },
     {
       "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": 28402.49634571118,
-      "required_pages_per_second_upper_estimate": 0.0021865027918348367
+      "required_token_positions_per_second": 28338.01960493809,
+      "required_pages_per_second_upper_estimate": 0.002181539193847082
     },
     {
       "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": 28402.49634571118,
-      "required_pages_per_second_upper_estimate": 0.35266394757330394
+      "required_token_positions_per_second": 28338.01960493809,
+      "required_pages_per_second_upper_estimate": 0.3518633622426722
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14914389588273025,
-    "producer_p95_seconds": 0.3874536770017585,
-    "gpu_update_seconds": 0.9734355622644024,
-    "producer_batches_per_second": 6.704934144849521,
-    "gpu_batches_per_second": 1.0272893643558731,
-    "producer_headroom": 6.526821339237384,
+    "producer_batch_seconds": 0.3832726171959848,
+    "producer_p95_seconds": 0.37597652999829734,
+    "gpu_update_seconds": 0.9756503942562786,
+    "producer_batches_per_second": 2.609108908734417,
+    "gpu_batches_per_second": 1.024957306312865,
+    "producer_headroom": 2.5455781354643032,
     "sustainable": true,
-    "buffer_seconds": 60.35300486039295,
+    "buffer_seconds": 60.49032444388928,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300258.5200801012,
+    "generation_bytes_per_second": 308497.2670563659,
     "upload_bytes_per_second": 5910628.800790494,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.42848874857214,
+    "estimated_full_shards_per_hour": 66.19633206146462,
     "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": [

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -337,3 +337,4 @@
 {"step": 118700, "loss": 0.6316186003386974, "lr": 0.0003, "grad_norm": 0.14607130050516742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17370383301, "tokens_per_second": 148738.52458228893, "data_wait_s": 1.0177995136473328e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 118800, "loss": 0.6226529106497765, "lr": 0.0003, "grad_norm": 0.14603244399728735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17384892783, "tokens_per_second": 148250.71523810277, "data_wait_s": 1.9462997443042696e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 118900, "loss": 0.5921273212879896, "lr": 0.0003, "grad_norm": 0.13395452390409357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17399398353, "tokens_per_second": 143621.63730805874, "data_wait_s": 1.3270997442305088e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 119000, "loss": 0.6057633049786091, "lr": 0.0003, "grad_norm": 0.15477933710314376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17414016155, "tokens_per_second": 148428.89229204145, "data_wait_s": 1.7439000657759607e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5047f9e7177725d8e1ad4378abec62a21ac46f050426122d8a3bf6a26f464628
-size 722865
+oid sha256:02bd8358e02bf547f2b603c84f8bca14944df51a4b6db06d33b4df76bba1c79c
+size 724983