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": 76948.88611645176,
-      "required_pages_per_second_upper_estimate": 0.005323502487550763
+      "required_token_positions_per_second": 75835.94029093218,
+      "required_pages_per_second_upper_estimate": 0.005246506312951219
     },
     {
       "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": 9618.61076455647,
-      "required_pages_per_second_upper_estimate": 0.004027111467686542
+      "required_token_positions_per_second": 9479.492536366522,
+      "required_pages_per_second_upper_estimate": 0.003968865570662368
     },
     {
       "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": 19237.22152911294,
-      "required_pages_per_second_upper_estimate": 0.004006605924866734
+      "required_token_positions_per_second": 18958.985072733045,
+      "required_pages_per_second_upper_estimate": 0.003948656608591588
     },
     {
       "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": 28855.83229366941,
-      "required_pages_per_second_upper_estimate": 0.001729863085700743
+      "required_token_positions_per_second": 28438.477609099566,
+      "required_pages_per_second_upper_estimate": 0.0017048433096244855
     },
     {
       "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": 28855.83229366941,
-      "required_pages_per_second_upper_estimate": 0.0022434140374402146
+      "required_token_positions_per_second": 28438.477609099566,
+      "required_pages_per_second_upper_estimate": 0.0022109665464641586
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28855.83229366941,
-      "required_pages_per_second_upper_estimate": 0.3701316336843987
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28438.477609099566,
+      "required_pages_per_second_upper_estimate": 0.36477825591128343
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10062771764035006,
-    "producer_p95_seconds": 0.3080614050013537,
-    "gpu_update_seconds": 0.9581425244859636,
-    "producer_batches_per_second": 9.937619807437791,
-    "gpu_batches_per_second": 1.043686063862464,
-    "producer_headroom": 9.521656129680162,
+    "producer_batch_seconds": 0.20390963517965588,
+    "producer_p95_seconds": 0.2665241059985419,
+    "gpu_update_seconds": 0.9722039407325154,
+    "producer_batches_per_second": 4.904133142697958,
+    "gpu_batches_per_second": 1.028590770005048,
+    "producer_headroom": 4.76781756720789,
     "sustainable": true,
-    "buffer_seconds": 60.362979042615706,
+    "buffer_seconds": 60.27664432541595,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 545463.5988112354,
-    "upload_bytes_per_second": 5927789.891576405,
+    "generation_bytes_per_second": 454673.46290830773,
+    "upload_bytes_per_second": 6031713.54210596,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.04379056217954,
+    "estimated_full_shards_per_hour": 97.56234088360713,
     "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": 1.0038025159992685,
-      "bytes": 2414763,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.1136625049985014,
       "bytes": 10469225,
@@ -562,6 +556,12 @@
       "bytes": 2649946,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2970311710014357,
+      "bytes": 11562430,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -261,3 +261,4 @@
 {"step": 423200, "loss": 0.5653585195541382, "lr": 0.0003, "grad_norm": 0.18599586864484288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61944825880, "tokens_per_second": 151543.49483212418, "data_wait_s": 1.2147000234108418e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 423300, "loss": 0.6064873933792114, "lr": 0.0003, "grad_norm": 0.15940243115982336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61959429259, "tokens_per_second": 151056.12213437274, "data_wait_s": 1.4637000276707113e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 423400, "loss": 0.5870299935340881, "lr": 0.0003, "grad_norm": 0.14720102250209122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61973986300, "tokens_per_second": 151409.76535355955, "data_wait_s": 8.121001883409917e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 423500, "loss": 0.5925978422164917, "lr": 0.0003, "grad_norm": 0.1639069225832975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61988663011, "tokens_per_second": 152502.25051917668, "data_wait_s": 1.3473996659740806e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f986cd7d1c2e4a2fd2404c349888661909eebd73f3d216cbec9d94276de113f2
-size 561114
+oid sha256:0abdf4d78d6511bfcb992b7261f8d59b475b36d02cc6567c039a5b4753cf24b4
+size 563232