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": 77273.16903939718,
-      "required_pages_per_second_upper_estimate": 0.005345937132600175
+      "required_token_positions_per_second": 77052.05607909137,
+      "required_pages_per_second_upper_estimate": 0.005330640024953466
     },
     {
       "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": 9659.146129924648,
-      "required_pages_per_second_upper_estimate": 0.004044082778691514
+      "required_token_positions_per_second": 9631.507009886422,
+      "required_pages_per_second_upper_estimate": 0.004032510856301967
     },
     {
       "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": 19318.292259849295,
-      "required_pages_per_second_upper_estimate": 0.004023490820100175
+      "required_token_positions_per_second": 19263.014019772843,
+      "required_pages_per_second_upper_estimate": 0.004011977820477472
     },
     {
       "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": 28977.43838977394,
-      "required_pages_per_second_upper_estimate": 0.0017371531854804521
+      "required_token_positions_per_second": 28894.521029659267,
+      "required_pages_per_second_upper_estimate": 0.0017321824163490556
     },
     {
       "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": 28977.43838977394,
-      "required_pages_per_second_upper_estimate": 0.0022528683765236537
+      "required_token_positions_per_second": 28894.521029659267,
+      "required_pages_per_second_upper_estimate": 0.002246421916489661
     },
     {
       "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": 28977.43838977394,
-      "required_pages_per_second_upper_estimate": 0.37169146611477455
+      "required_token_positions_per_second": 28894.521029659267,
+      "required_pages_per_second_upper_estimate": 0.3706278912489484
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09771086398440332,
-    "producer_p95_seconds": 0.19127077199937048,
-    "gpu_update_seconds": 0.9541216041289869,
-    "producer_batches_per_second": 10.23427650951506,
-    "gpu_batches_per_second": 1.048084432500504,
-    "producer_headroom": 9.764744320358119,
+    "producer_batch_seconds": 0.3612854624609838,
+    "producer_p95_seconds": 0.1833698740001637,
+    "gpu_update_seconds": 0.9568596057231835,
+    "producer_batches_per_second": 2.7678943768959225,
+    "gpu_batches_per_second": 1.0450853960380233,
+    "producer_headroom": 2.648486322160049,
     "sustainable": true,
-    "buffer_seconds": 60.10966106012618,
+    "buffer_seconds": 60.28215516056056,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337768.9967947278,
-    "upload_bytes_per_second": 4149365.5205726246,
+    "generation_bytes_per_second": 340789.8806861911,
+    "upload_bytes_per_second": 4127433.343550122,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.4773638523233,
+    "estimated_full_shards_per_hour": 73.12557521285342,
     "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": [
@@ -418,6 +418,12 @@
       "bytes": 12346633,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9011206420000235,
+      "bytes": 2301339,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -65,3 +65,4 @@
 {"step": 531500, "loss": 0.5958731174468994, "lr": 0.0003, "grad_norm": 0.19584721023642038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77794232968, "tokens_per_second": 153315.87980441257, "data_wait_s": 1.239099947270006e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 531600, "loss": 0.5742305517196655, "lr": 0.0003, "grad_norm": 0.17469050618035542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77808923030, "tokens_per_second": 153622.97289850537, "data_wait_s": 1.7005999325192533e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 531700, "loss": 0.601482093334198, "lr": 0.0003, "grad_norm": 0.16568858733734793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77823771755, "tokens_per_second": 155227.28266635895, "data_wait_s": 6.833999577793293e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 531800, "loss": 0.6045354008674622, "lr": 0.0003, "grad_norm": 0.18553475762520746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77838537453, "tokens_per_second": 154608.2630370454, "data_wait_s": 1.0071999895444605e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:58ebe900d3ef98ae66b35ba65e54a749907fc54d55d74aa524377ea2b89591d8
-size 142854
+oid sha256:a978f4e74455ac4b35121cde02aa5aee1c981487347b4a6e1a142de1c389ec95
+size 144972