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": 75090.18674026288,
-      "required_pages_per_second_upper_estimate": 0.006508627072307632
+      "required_token_positions_per_second": 75472.88389841687,
+      "required_pages_per_second_upper_estimate": 0.0065417982920393395
     },
     {
       "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": 9386.27334253286,
-      "required_pages_per_second_upper_estimate": 0.003744324773628874
+      "required_token_positions_per_second": 9434.110487302109,
+      "required_pages_per_second_upper_estimate": 0.0037634077259063783
     },
     {
       "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": 18772.54668506572,
-      "required_pages_per_second_upper_estimate": 0.004342698290052365
+      "required_token_positions_per_second": 18868.220974604217,
+      "required_pages_per_second_upper_estimate": 0.004364830853127109
     },
     {
       "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": 28158.82002759858,
-      "required_pages_per_second_upper_estimate": 0.001714724401769856
+      "required_token_positions_per_second": 28302.331461906328,
+      "required_pages_per_second_upper_estimate": 0.0017234634951728929
     },
     {
       "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": 28158.82002759858,
-      "required_pages_per_second_upper_estimate": 0.0021677439143273076
+      "required_token_positions_per_second": 28302.331461906328,
+      "required_pages_per_second_upper_estimate": 0.0021787918218053944
     },
     {
       "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": 28158.82002759858,
-      "required_pages_per_second_upper_estimate": 0.34963830323452055
+      "required_token_positions_per_second": 28302.331461906328,
+      "required_pages_per_second_upper_estimate": 0.35142023494674907
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.08711004778135134,
-    "producer_p95_seconds": 0.24109585899714148,
-    "gpu_update_seconds": 0.9818593241088255,
-    "producer_batches_per_second": 11.479731965134814,
-    "gpu_batches_per_second": 1.018475840118583,
-    "producer_headroom": 11.271481868237748,
+    "producer_batch_seconds": 0.11705174973420185,
+    "producer_p95_seconds": 0.3109116340019682,
+    "gpu_update_seconds": 0.9768806515891799,
+    "producer_batches_per_second": 8.543229830145851,
+    "gpu_batches_per_second": 1.0236665025284406,
+    "producer_headroom": 8.345715923149,
     "sustainable": true,
-    "buffer_seconds": 60.87527809474718,
+    "buffer_seconds": 60.566600398529154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369594.8922087617,
-    "upload_bytes_per_second": 6345393.411090532,
+    "generation_bytes_per_second": 354991.44489502657,
+    "upload_bytes_per_second": 6424576.631309033,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.3064601392473,
+    "estimated_full_shards_per_hour": 76.17290029657457,
     "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": 0.8243736469994474,
-      "bytes": 1000268,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.0672549959999742,
       "bytes": 7538580,
@@ -562,6 +556,12 @@
       "bytes": 5747516,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.016742787000112,
+      "bytes": 7467282,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -195,3 +195,4 @@
 {"step": 59500, "loss": 0.6615736372768879, "lr": 0.0003, "grad_norm": 0.13941057722251005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8708364498, "tokens_per_second": 147979.99047645446, "data_wait_s": 8.458999218419194e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 59600, "loss": 0.5707687921822071, "lr": 0.0003, "grad_norm": 0.14786385360195126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8722874790, "tokens_per_second": 141383.39511530905, "data_wait_s": 8.326001989189535e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 59700, "loss": 0.6407595612108707, "lr": 0.0003, "grad_norm": 0.13687393807649878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8737571646, "tokens_per_second": 149532.1105406038, "data_wait_s": 1.1885000276379287e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 59800, "loss": 0.6514598056674004, "lr": 0.0003, "grad_norm": 0.13604662731343992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8752313117, "tokens_per_second": 150564.75548546857, "data_wait_s": 1.812400296330452e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b446fff40a230efda7be113db198126a7bcc7281c1288349715078e247cbba73
-size 419760
+oid sha256:b7011f7831f29890cd02987aa650b152726e0014f39e4dd056a1d10ad8960a79
+size 421878