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": 75510.1334509874,
-      "required_pages_per_second_upper_estimate": 0.006545026988848025
+      "required_token_positions_per_second": 75611.87861269429,
+      "required_pages_per_second_upper_estimate": 0.006553846001594024
     },
     {
       "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": 9438.766681373425,
-      "required_pages_per_second_upper_estimate": 0.0037652651513377313
+      "required_token_positions_per_second": 9451.484826586786,
+      "required_pages_per_second_upper_estimate": 0.003770338609616557
     },
     {
       "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": 18877.53336274685,
-      "required_pages_per_second_upper_estimate": 0.0043669851102314736
+      "required_token_positions_per_second": 18902.96965317357,
+      "required_pages_per_second_upper_estimate": 0.004372869348358273
     },
     {
       "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": 28316.30004412028,
-      "required_pages_per_second_upper_estimate": 0.0017243141085422355
+      "required_token_positions_per_second": 28354.45447976036,
+      "required_pages_per_second_upper_estimate": 0.0017266375134918634
     },
     {
       "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": 28316.30004412028,
-      "required_pages_per_second_upper_estimate": 0.0021798671619317348
+      "required_token_positions_per_second": 28354.45447976036,
+      "required_pages_per_second_upper_estimate": 0.0021828043960055456
     },
     {
       "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": 28316.30004412028,
-      "required_pages_per_second_upper_estimate": 0.35159367798800895
+      "required_token_positions_per_second": 28354.45447976036,
+      "required_pages_per_second_upper_estimate": 0.35206742838397703
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09809951397657812,
-    "producer_p95_seconds": 0.19824715699996887,
-    "gpu_update_seconds": 0.9763987511405451,
-    "producer_batches_per_second": 10.193730421934163,
-    "gpu_batches_per_second": 1.0241717319198596,
-    "producer_headroom": 9.953145653439899,
+    "producer_batch_seconds": 0.07321604433597884,
+    "producer_p95_seconds": 0.2082474610006102,
+    "gpu_update_seconds": 0.9750848855066271,
+    "producer_batches_per_second": 13.658208512483016,
+    "gpu_batches_per_second": 1.025551738995962,
+    "producer_headroom": 13.31791268362014,
     "sustainable": true,
-    "buffer_seconds": 60.53672257071379,
+    "buffer_seconds": 60.455262901410876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337106.65336882,
-    "upload_bytes_per_second": 3462518.697251743,
+    "generation_bytes_per_second": 380796.0303893367,
+    "upload_bytes_per_second": 3441806.4113709875,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.33524037168932,
+    "estimated_full_shards_per_hour": 81.70996364364696,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 5987959,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.322158891000072,
+      "bytes": 2479484,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -64,3 +64,4 @@
 {"step": 243500, "loss": 0.5820450047031045, "lr": 0.0003, "grad_norm": 0.146134232664598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35640810545, "tokens_per_second": 150072.78225572026, "data_wait_s": 9.190000128000975e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 243600, "loss": 0.5845957063138485, "lr": 0.0003, "grad_norm": 0.15045103453201752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35655504986, "tokens_per_second": 149962.21649101932, "data_wait_s": 3.174199991917703e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 243700, "loss": 0.6125733498483896, "lr": 0.0003, "grad_norm": 0.1604887896686748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35670226012, "tokens_per_second": 150125.01255507447, "data_wait_s": 1.1823000022559427e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 243800, "loss": 0.6316460017114878, "lr": 0.0003, "grad_norm": 0.17855176982115806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35685005397, "tokens_per_second": 150819.9502725219, "data_wait_s": 1.1596000149438623e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:693ced4520cf9bd1ee6f10d3825def0378aa1b1830b3634f35d94b04b4b0f3fb
-size 140736
+oid sha256:bf3c51a30faa1c0fcff302e17dc207a677c7f813bb50048fd4d05bd677128858
+size 142854