Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77338.202067346,
-      "required_pages_per_second_upper_estimate": 0.0053504362683710885
+      "required_token_positions_per_second": 77205.69661380946,
+      "required_pages_per_second_upper_estimate": 0.0053412692336404196
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9667.27525841825,
-      "required_pages_per_second_upper_estimate": 0.004047486275036279
+      "required_token_positions_per_second": 9650.712076726182,
+      "required_pages_per_second_upper_estimate": 0.004040551616740375
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19334.5505168365,
-      "required_pages_per_second_upper_estimate": 0.00402687698627154
+      "required_token_positions_per_second": 19301.424153452364,
+      "required_pages_per_second_upper_estimate": 0.004019977638379574
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29001.82577525475,
-      "required_pages_per_second_upper_estimate": 0.0017386151720026454
+      "required_token_positions_per_second": 28952.136230178545,
+      "required_pages_per_second_upper_estimate": 0.001735636359646869
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29001.82577525475,
-      "required_pages_per_second_upper_estimate": 0.0022547643884760146
+      "required_token_positions_per_second": 28952.136230178545,
+      "required_pages_per_second_upper_estimate": 0.00225090124490755
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29001.82577525475,
-      "required_pages_per_second_upper_estimate": 0.372004281310588
+      "required_token_positions_per_second": 28952.136230178545,
+      "required_pages_per_second_upper_estimate": 0.3713669171788272
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13410358990626037,
-    "producer_p95_seconds": 0.321670900004392,
-    "gpu_update_seconds": 0.953319291490611,
-    "producer_batches_per_second": 7.456921926542079,
-    "gpu_batches_per_second": 1.0489664993943415,
-    "producer_headroom": 7.108827527711897,
+    "producer_batch_seconds": 0.08611524378136437,
+    "producer_p95_seconds": 0.29751900900009787,
+    "gpu_update_seconds": 0.9549554402545548,
+    "producer_batches_per_second": 11.612345922620536,
+    "gpu_batches_per_second": 1.0471692791586569,
+    "producer_headroom": 11.089272912924278,
     "sustainable": true,
-    "buffer_seconds": 60.0591153639085,
+    "buffer_seconds": 60.16219273603695,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377839.804014771,
-    "upload_bytes_per_second": 5757896.934755872,
+    "generation_bytes_per_second": 362245.4411740977,
+    "upload_bytes_per_second": 5705183.1239266,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.0756262810931,
+    "estimated_full_shards_per_hour": 77.72943903367232,
     "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": 1.038082987000962,
-      "bytes": 8819669,
-      "pages": 48,
-      "shards": 3
-    },
     {
       "seconds": 1.3574339080005302,
       "bytes": 10377609,
@@ -562,6 +556,12 @@
       "bytes": 1229989,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.4636801879969425,
+      "bytes": 7357161,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -351,3 +351,4 @@
 {"step": 720100, "loss": 0.5303980708122253, "lr": 8e-05, "grad_norm": 0.20666587655957588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105402386187, "tokens_per_second": 141301.18391503618, "data_wait_s": 1.5104997146409005e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720200, "loss": 0.5344586372375488, "lr": 8e-05, "grad_norm": 0.22277006736838195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105416991235, "tokens_per_second": 151680.44329242178, "data_wait_s": 1.6111000149976462e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720300, "loss": 0.5282149910926819, "lr": 8e-05, "grad_norm": 0.21542854874343864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105431760929, "tokens_per_second": 153713.69739467345, "data_wait_s": 1.4101999113336205e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 720400, "loss": 0.5309709310531616, "lr": 8e-05, "grad_norm": 0.2127561535188853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105446489981, "tokens_per_second": 153689.68463523037, "data_wait_s": 1.5589997929055244e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6fe8827739a865532f8fda0d3cfbf1ba290b74b8397db3ca8852937508b058b7
-size 753300
+oid sha256:5eeea5b171ad8703e099ac715c5d99c99b4ff36d178190d7e22959d9c2702102
+size 755418