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": 77290.57446864402,
-      "required_pages_per_second_upper_estimate": 0.0053471412805816305
+      "required_token_positions_per_second": 76292.35472245458,
+      "required_pages_per_second_upper_estimate": 0.005278082122351325
     },
     {
       "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": 9661.321808580502,
-      "required_pages_per_second_upper_estimate": 0.004044993689911383
+      "required_token_positions_per_second": 9536.544340306822,
+      "required_pages_per_second_upper_estimate": 0.003992751969595029
     },
     {
       "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": 19322.643617161004,
-      "required_pages_per_second_upper_estimate": 0.004024397093075194
+      "required_token_positions_per_second": 19073.088680613644,
+      "required_pages_per_second_upper_estimate": 0.003972421380998623
     },
     {
       "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": 28983.965425741506,
-      "required_pages_per_second_upper_estimate": 0.001737544471320501
+      "required_token_positions_per_second": 28609.633020920468,
+      "required_pages_per_second_upper_estimate": 0.0017151038152239668
     },
     {
       "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": 28983.965425741506,
-      "required_pages_per_second_upper_estimate": 0.0022533758248607346
+      "required_token_positions_per_second": 28609.633020920468,
+      "required_pages_per_second_upper_estimate": 0.002224273126900139
     },
     {
       "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": 28983.965425741506,
-      "required_pages_per_second_upper_estimate": 0.3717751879239813
+      "required_token_positions_per_second": 28609.633020920468,
+      "required_pages_per_second_upper_estimate": 0.36697365376175867
     }
   ],
   "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": 0.10474604682805477,
-    "producer_p95_seconds": 0.19070651599940902,
-    "gpu_update_seconds": 0.9539067409818603,
-    "producer_batches_per_second": 9.546899670987525,
-    "gpu_batches_per_second": 1.048320508743544,
-    "producer_headroom": 9.106851951632505,
+    "producer_batch_seconds": 0.6593967394140066,
+    "producer_p95_seconds": 0.18325108800127055,
+    "gpu_update_seconds": 0.9663877890283569,
+    "producer_batches_per_second": 1.5165376778912816,
+    "gpu_batches_per_second": 1.0347812869256534,
+    "producer_headroom": 1.465563493515554,
     "sustainable": true,
-    "buffer_seconds": 60.096124681857205,
+    "buffer_seconds": 60.88243070878648,
     "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": 256092.91438012492,
-    "upload_bytes_per_second": 5768655.250378937,
+    "generation_bytes_per_second": 224681.9476634356,
+    "upload_bytes_per_second": 5903167.997206218,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.951577888038734,
+    "estimated_full_shards_per_hour": 48.21151564051915,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 2610307,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.183507477000603,
+      "bytes": 15672429,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -72,3 +72,4 @@
 {"step": 492200, "loss": 0.5845812559127808, "lr": 0.0003, "grad_norm": 0.14824155212671003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72041630681, "tokens_per_second": 153710.4138390692, "data_wait_s": 9.390998457092792e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 492300, "loss": 0.5877999067306519, "lr": 0.0003, "grad_norm": 0.15603947595963416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72056303487, "tokens_per_second": 152884.0486768965, "data_wait_s": 8.569000783609226e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 492400, "loss": 0.5546084642410278, "lr": 0.0003, "grad_norm": 0.14702425554910908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72070868488, "tokens_per_second": 145653.51486198275, "data_wait_s": 1.738399987516459e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 492500, "loss": 0.5923275351524353, "lr": 0.0003, "grad_norm": 0.14910898405207795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72085369174, "tokens_per_second": 151148.39238931972, "data_wait_s": 1.1112999345641583e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8043f48d6a1fe151d932c13f502a78bfcf46d71fddde68257292a24063dad25b
-size 157680
+oid sha256:847bab0eae0d5d1b3d992f46dc3b9279057eae237f66dd79e7de2a6b9b0fa756
+size 159798