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": 75829.46412435843,
-      "required_pages_per_second_upper_estimate": 0.006572705762279634
+      "required_token_positions_per_second": 74789.22219739005,
+      "required_pages_per_second_upper_estimate": 0.0064825402285190745
     },
     {
       "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": 9478.683015544804,
-      "required_pages_per_second_upper_estimate": 0.003781188373841074
+      "required_token_positions_per_second": 9348.652774673756,
+      "required_pages_per_second_upper_estimate": 0.0037293173666322625
     },
     {
       "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": 18957.36603108961,
-      "required_pages_per_second_upper_estimate": 0.004385452993045595
+      "required_token_positions_per_second": 18697.30554934751,
+      "required_pages_per_second_upper_estimate": 0.004325292577502718
     },
     {
       "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": 28436.049046634413,
-      "required_pages_per_second_upper_estimate": 0.0017316061945208856
+      "required_token_positions_per_second": 28045.958324021267,
+      "required_pages_per_second_upper_estimate": 0.0017078517161615934
     },
     {
       "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": 28436.049046634413,
-      "required_pages_per_second_upper_estimate": 0.0021890857716317276
+      "required_token_positions_per_second": 28045.958324021267,
+      "required_pages_per_second_upper_estimate": 0.0021590555079647345
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28436.049046634413,
-      "required_pages_per_second_upper_estimate": 0.35308055982510417
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28045.958324021267,
+      "required_pages_per_second_upper_estimate": 0.3482369385999139
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09967690590656275,
-    "producer_p95_seconds": 0.3118667060007283,
-    "gpu_update_seconds": 0.9722869711842869,
-    "producer_batches_per_second": 10.03241413750745,
-    "gpu_batches_per_second": 1.028502931374219,
-    "producer_headroom": 9.754385555423537,
+    "producer_batch_seconds": 0.42701781460147004,
+    "producer_p95_seconds": 0.3386782650013629,
+    "gpu_update_seconds": 0.9858104929265185,
+    "producer_batches_per_second": 2.341822672979783,
+    "gpu_batches_per_second": 1.0143937472519267,
+    "producer_headroom": 2.308593363596697,
     "sustainable": true,
-    "buffer_seconds": 60.28179221342579,
+    "buffer_seconds": 60.13444006851763,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308833.90848106355,
-    "upload_bytes_per_second": 5690670.628727696,
+    "generation_bytes_per_second": 297301.97635776654,
+    "upload_bytes_per_second": 5711591.078716606,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.26856747459345,
+    "estimated_full_shards_per_hour": 63.79408329057453,
     "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": 1.2244086760001665,
-      "bytes": 5469696,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.178703661000327,
       "bytes": 7809750,
@@ -562,6 +556,12 @@
       "bytes": 11728601,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 0.9978417950005678,
+      "bytes": 5753379,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 304700, "loss": 0.5947480089962482, "lr": 0.0003, "grad_norm": 0.14075108484866727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44600248529, "tokens_per_second": 150838.17199795527, "data_wait_s": 1.058699854183942e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 304800, "loss": 0.5849448200315237, "lr": 0.0003, "grad_norm": 0.16570698714419854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44614851586, "tokens_per_second": 143666.1496722523, "data_wait_s": 1.3000997569179162e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 304900, "loss": 0.554118636995554, "lr": 0.0003, "grad_norm": 0.14330116719255986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44629402983, "tokens_per_second": 149150.65887491562, "data_wait_s": 1.2025000614812598e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 305000, "loss": 0.6317330747842789, "lr": 0.0003, "grad_norm": 0.15201925927561333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44643840707, "tokens_per_second": 147406.7625548003, "data_wait_s": 1.454499943065457e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1bdb2a2ec6029b5c31203a4b1cce32abc3f71b3b91605cae3712ee5b8188a034
-size 526443
+oid sha256:175fe345d2a6b7bd229c36a87ca3d942f67acb84a9ad3f306578e1e951f8d42c
+size 528561