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": 74523.39750279277,
-      "required_pages_per_second_upper_estimate": 0.006459499217717911
+      "required_token_positions_per_second": 75040.63641009977,
+      "required_pages_per_second_upper_estimate": 0.0065043321752732575
     },
     {
       "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": 9315.424687849096,
-      "required_pages_per_second_upper_estimate": 0.003716062185993735
+      "required_token_positions_per_second": 9380.079551262472,
+      "required_pages_per_second_upper_estimate": 0.0037418539776856836
     },
     {
       "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": 18630.84937569819,
-      "required_pages_per_second_upper_estimate": 0.004309919111317658
+      "required_token_positions_per_second": 18760.159102524944,
+      "required_pages_per_second_upper_estimate": 0.0043398326408455624
     },
     {
       "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": 27946.274063547284,
-      "required_pages_per_second_upper_estimate": 0.0017017814677016227
+      "required_token_positions_per_second": 28140.238653787415,
+      "required_pages_per_second_upper_estimate": 0.0017135928935936903
     },
     {
       "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": 27946.274063547284,
-      "required_pages_per_second_upper_estimate": 0.0021513815376497532
+      "required_token_positions_per_second": 28140.238653787415,
+      "required_pages_per_second_upper_estimate": 0.0021663134687347893
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27946.274063547284,
-      "required_pages_per_second_upper_estimate": 0.3469991937065856
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28140.238653787415,
+      "required_pages_per_second_upper_estimate": 0.34940758475964356
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0539737678437433,
-    "producer_p95_seconds": 0.1661246610001399,
-    "gpu_update_seconds": 0.9893268754586376,
-    "producer_batches_per_second": 18.52751882905505,
-    "gpu_batches_per_second": 1.010788269080848,
-    "producer_headroom": 18.32977231315011,
+    "producer_batch_seconds": 0.2509253884218907,
+    "producer_p95_seconds": 0.1730511679998017,
+    "gpu_update_seconds": 0.9825076588779689,
+    "producer_batches_per_second": 3.985248389129364,
+    "gpu_batches_per_second": 1.0178037707533063,
+    "producer_headroom": 3.9155370648506884,
     "sustainable": true,
-    "buffer_seconds": 60.348939402976896,
+    "buffer_seconds": 60.91547485043407,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375920.2819864508,
-    "upload_bytes_per_second": 5548075.735603926,
+    "generation_bytes_per_second": 368399.1148457734,
+    "upload_bytes_per_second": 5743694.839952308,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.66374153800146,
+    "estimated_full_shards_per_hour": 79.04987415342237,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 6894198,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.282220471999608,
+      "bytes": 9425496,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -39,3 +39,4 @@
 {"step": 198900, "loss": 0.5834132824093103, "lr": 0.0003, "grad_norm": 0.15648794412353406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29111204742, "tokens_per_second": 147019.28082561592, "data_wait_s": 2.321600004506763e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 199000, "loss": 0.5890722051262856, "lr": 0.0003, "grad_norm": 0.14006582485757682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29125878764, "tokens_per_second": 148295.2887275073, "data_wait_s": 2.1120000383234583e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 199100, "loss": 0.5936685418710113, "lr": 0.0003, "grad_norm": 0.1579531949293927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29140523105, "tokens_per_second": 147785.8147173087, "data_wait_s": 1.7545000446261838e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 199200, "loss": 0.5824197605252266, "lr": 0.0003, "grad_norm": 0.13886453172844473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29154937790, "tokens_per_second": 145347.49475511748, "data_wait_s": 1.88409994734684e-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 sha256:6df4b4b640a5bc4ceda8117ce64377f1cfd1efbfc8cc6ebc07bab93be78d7792
-size 87003
+oid sha256:aef67309ab361e55e80087f42c56c4a7b296f4a314b8381e173e3d9202a3fa00
+size 89121