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": 75420.37933958005,
-      "required_pages_per_second_upper_estimate": 0.0065372473299509515
+      "required_token_positions_per_second": 74793.96698767327,
+      "required_pages_per_second_upper_estimate": 0.006482951494915263
     },
     {
       "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": 9427.547417447506,
-      "required_pages_per_second_upper_estimate": 0.003760789619214738
+      "required_token_positions_per_second": 9349.245873459158,
+      "required_pages_per_second_upper_estimate": 0.0037295539626053765
     },
     {
       "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": 18855.09483489501,
-      "required_pages_per_second_upper_estimate": 0.004361794351717553
+      "required_token_positions_per_second": 18698.491746918316,
+      "required_pages_per_second_upper_estimate": 0.00432556698343436
     },
     {
       "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": 28282.642252342514,
-      "required_pages_per_second_upper_estimate": 0.0017222645256117556
+      "required_token_positions_per_second": 28047.737620377477,
+      "required_pages_per_second_upper_estimate": 0.00170796006597444
     },
     {
       "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": 28282.642252342514,
-      "required_pages_per_second_upper_estimate": 0.0021772760919499512
+      "required_token_positions_per_second": 28047.737620377477,
+      "required_pages_per_second_upper_estimate": 0.0021591924831236434
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28282.642252342514,
-      "required_pages_per_second_upper_estimate": 0.35117576085951197
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28047.737620377477,
+      "required_pages_per_second_upper_estimate": 0.3482590315057362
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10818191460148796,
-    "producer_p95_seconds": 0.3064405849981995,
-    "gpu_update_seconds": 0.9775607156262088,
-    "producer_batches_per_second": 9.243689240329324,
-    "gpu_batches_per_second": 1.0229543638723422,
-    "producer_headroom": 9.03626746880262,
+    "producer_batch_seconds": 0.101036265132592,
+    "producer_p95_seconds": 0.34567069499826175,
+    "gpu_update_seconds": 0.9857479549406846,
+    "producer_batches_per_second": 9.8974363184118,
+    "gpu_batches_per_second": 1.0144581025888844,
+    "producer_headroom": 9.756377610030091,
     "sustainable": true,
-    "buffer_seconds": 60.60876436882495,
+    "buffer_seconds": 60.13062525138176,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377183.1623781689,
+    "generation_bytes_per_second": 376206.6096165109,
     "upload_bytes_per_second": 2733627.039630479,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.93472627171326,
+    "estimated_full_shards_per_hour": 80.72518078204628,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -259,3 +259,4 @@
 {"step": 220900, "loss": 0.5657839942723513, "lr": 0.0003, "grad_norm": 0.13758471348440585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32333370010, "tokens_per_second": 150897.88842711048, "data_wait_s": 1.410699769621715e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 221000, "loss": 0.6313675027340651, "lr": 0.0003, "grad_norm": 0.16740018193347608, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32347900388, "tokens_per_second": 146695.52538510278, "data_wait_s": 1.0517000191612169e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 221100, "loss": 0.5796709470450878, "lr": 0.0003, "grad_norm": 0.13842052592317736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32362371517, "tokens_per_second": 146255.55999118107, "data_wait_s": 1.6242000128841028e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 221200, "loss": 0.5717740878462791, "lr": 0.0003, "grad_norm": 0.13649338577095727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32376948957, "tokens_per_second": 148279.05683170797, "data_wait_s": 1.471799987484701e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ece92bd1981c3e0a658962ddc4a9955504121a66bf10d769ce57968a66c3cd1
-size 556878
+oid sha256:9a055c10ac8a1037a03f6c7e7abca7e80c77a8d965120cc0f34ced3376cfe4ce
+size 558996