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": 75274.09468107914,
-      "required_pages_per_second_upper_estimate": 0.00652456774650717
+      "required_token_positions_per_second": 75333.12188192015,
+      "required_pages_per_second_upper_estimate": 0.006529684074672996
     },
     {
       "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": 9409.261835134892,
-      "required_pages_per_second_upper_estimate": 0.003753495227036418
+      "required_token_positions_per_second": 9416.640235240018,
+      "required_pages_per_second_upper_estimate": 0.003756438581155265
     },
     {
       "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": 18818.523670269784,
-      "required_pages_per_second_upper_estimate": 0.004353334256412021
+      "required_token_positions_per_second": 18833.280470480036,
+      "required_pages_per_second_upper_estimate": 0.00435674798242985
     },
     {
       "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": 28227.785505404674,
-      "required_pages_per_second_upper_estimate": 0.0017189240375343493
+      "required_token_positions_per_second": 28249.920705720055,
+      "required_pages_per_second_upper_estimate": 0.0017202719550991352
     },
     {
       "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": 28227.785505404674,
-      "required_pages_per_second_upper_estimate": 0.002173053067717482
+      "required_token_positions_per_second": 28249.920705720055,
+      "required_pages_per_second_upper_estimate": 0.002174757096711916
     },
     {
       "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": 28227.785505404674,
-      "required_pages_per_second_upper_estimate": 0.35049462365626577
+      "required_token_positions_per_second": 28249.920705720055,
+      "required_pages_per_second_upper_estimate": 0.35076946876243287
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.4423481663827964,
-    "producer_p95_seconds": 0.15610950300015247,
-    "gpu_update_seconds": 0.9794604679388093,
-    "producer_batches_per_second": 2.2606626996496386,
-    "gpu_batches_per_second": 1.020970251208213,
-    "producer_headroom": 2.214229745650647,
+    "producer_batch_seconds": 0.08265488513282904,
+    "producer_p95_seconds": 0.16153355100004774,
+    "gpu_update_seconds": 0.9786930125577953,
+    "producer_batches_per_second": 12.098498454059529,
+    "gpu_batches_per_second": 1.021770858858509,
+    "producer_headroom": 11.840715899429348,
     "sustainable": true,
-    "buffer_seconds": 60.72654901220617,
+    "buffer_seconds": 60.67896677858331,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278842.9472913221,
-    "upload_bytes_per_second": 4485651.663916101,
+    "generation_bytes_per_second": 279807.73113581986,
+    "upload_bytes_per_second": 4550910.816382349,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.833205357120015,
+    "estimated_full_shards_per_hour": 60.04022551113078,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 856023,
       "pages": 24,
       "shards": 1
+    },
+    {
+      "seconds": 1.8453115359998264,
+      "bytes": 9879354,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -21,3 +21,4 @@
 {"step": 317100, "loss": 0.5828649736940861, "lr": 0.0003, "grad_norm": 0.1620742490346258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46415209341, "tokens_per_second": 149288.2822589277, "data_wait_s": 1.3791999663226306e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 317200, "loss": 0.604051448404789, "lr": 0.0003, "grad_norm": 0.1369729481549036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46429715538, "tokens_per_second": 121843.6044977407, "data_wait_s": 1.4817000192124397e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 317300, "loss": 0.6179237812757492, "lr": 0.0003, "grad_norm": 0.16623904177407978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46444267414, "tokens_per_second": 147122.58520179737, "data_wait_s": 1.336800005447003e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 317400, "loss": 0.5875317845493555, "lr": 0.0003, "grad_norm": 0.1388448377576693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46458853229, "tokens_per_second": 147464.9528725937, "data_wait_s": 1.1488999916764442e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:89242ac3e26ebbe05b79b676a5fcbadad1ef51739d2df0a5f36ca90bc4fb77e2
-size 48879
+oid sha256:61cbec2148715aad7c9e4b1879528ce698cd99e525e948787ce8d007138a82c2
+size 50997