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": 77134.85176123504,
-      "required_pages_per_second_upper_estimate": 0.005336368022356616
+      "required_token_positions_per_second": 76948.88611645176,
+      "required_pages_per_second_upper_estimate": 0.005323502487550763
     },
     {
       "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": 9641.85647015438,
-      "required_pages_per_second_upper_estimate": 0.004036843959194855
+      "required_token_positions_per_second": 9618.61076455647,
+      "required_pages_per_second_upper_estimate": 0.004027111467686542
     },
     {
       "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": 19283.71294030876,
-      "required_pages_per_second_upper_estimate": 0.004016288859757861
+      "required_token_positions_per_second": 19237.22152911294,
+      "required_pages_per_second_upper_estimate": 0.004006605924866734
     },
     {
       "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": 28925.569410463144,
-      "required_pages_per_second_upper_estimate": 0.0017340437193701156
+      "required_token_positions_per_second": 28855.83229366941,
+      "required_pages_per_second_upper_estimate": 0.001729863085700743
     },
     {
       "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": 28925.569410463144,
-      "required_pages_per_second_upper_estimate": 0.0022488357915297678
+      "required_token_positions_per_second": 28855.83229366941,
+      "required_pages_per_second_upper_estimate": 0.0022434140374402146
     },
     {
       "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": 28925.569410463144,
-      "required_pages_per_second_upper_estimate": 0.37102614654074656
+      "required_token_positions_per_second": 28855.83229366941,
+      "required_pages_per_second_upper_estimate": 0.3701316336843987
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.12662019499995836,
-    "producer_p95_seconds": 0.28695402199809905,
-    "gpu_update_seconds": 0.9558325233866957,
-    "producer_batches_per_second": 7.897634338663978,
-    "gpu_batches_per_second": 1.0462083843483485,
-    "producer_headroom": 7.5488157587106075,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.217448973361826,
+    "buffer_seconds": 60.362979042615706,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 538103.6359360487,
-    "upload_bytes_per_second": 5973852.393594249,
+    "generation_bytes_per_second": 545463.5988112354,
+    "upload_bytes_per_second": 5927789.891576405,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.4645138603315,
+    "estimated_full_shards_per_hour": 117.04379056217954,
     "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": 0.9641569820014411,
-      "bytes": 6035455,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.0038025159992685,
       "bytes": 2414763,
@@ -562,6 +556,12 @@
       "bytes": 5757385,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9449253849998058,
+      "bytes": 2649946,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -260,3 +260,4 @@
 {"step": 423100, "loss": 0.553971529006958, "lr": 0.0003, "grad_norm": 0.16677586949778886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61930193436, "tokens_per_second": 150565.42364819525, "data_wait_s": 6.823000148870051e-06, "gpu_peak_gb": 89.37861251831055}
 {"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}

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 sha256:6908f576b5f3388607f0574abd8d66371eac60faa440d924ceb6d3c0d7e9185e
-size 558996
+oid sha256:f986cd7d1c2e4a2fd2404c349888661909eebd73f3d216cbec9d94276de113f2
+size 561114