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": 74911.41872948791,
-      "required_pages_per_second_upper_estimate": 0.006493131914217038
+      "required_token_positions_per_second": 75053.56339596701,
+      "required_pages_per_second_upper_estimate": 0.006505452653645083
     },
     {
       "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": 9363.927341185989,
-      "required_pages_per_second_upper_estimate": 0.003735410619589113
+      "required_token_positions_per_second": 9381.695424495876,
+      "required_pages_per_second_upper_estimate": 0.003742498573677946
     },
     {
       "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": 18727.854682371977,
-      "required_pages_per_second_upper_estimate": 0.00433235958178155
+      "required_token_positions_per_second": 18763.390848991752,
+      "required_pages_per_second_upper_estimate": 0.004340580248513863
     },
     {
       "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": 28091.782023557967,
-      "required_pages_per_second_upper_estimate": 0.0017106421390450076
+      "required_token_positions_per_second": 28145.086273487632,
+      "required_pages_per_second_upper_estimate": 0.0017138880881999385
     },
     {
       "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": 28091.782023557967,
-      "required_pages_per_second_upper_estimate": 0.002162583143203189
+      "required_token_positions_per_second": 28145.086273487632,
+      "required_pages_per_second_upper_estimate": 0.0021666866519183823
     },
     {
       "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": 28091.782023557967,
-      "required_pages_per_second_upper_estimate": 0.34880591558610285
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28145.086273487632,
+      "required_pages_per_second_upper_estimate": 0.34946777597238077
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5247172462733829,
-    "producer_p95_seconds": 0.38856797499465756,
-    "gpu_update_seconds": 0.9842024253503815,
-    "producer_batches_per_second": 1.9057883214286233,
-    "gpu_batches_per_second": 1.0160511437918824,
-    "producer_headroom": 1.8756814881544837,
+    "producer_batch_seconds": 0.1342781361171319,
+    "producer_p95_seconds": 0.34644466499594273,
+    "gpu_update_seconds": 0.9823384348991717,
+    "producer_batches_per_second": 7.447228781368337,
+    "gpu_batches_per_second": 1.0179791042204727,
+    "producer_headroom": 7.315699065425439,
     "sustainable": true,
-    "buffer_seconds": 60.03634794637327,
+    "buffer_seconds": 60.904982963748644,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298099.9229165502,
-    "upload_bytes_per_second": 5950221.628526375,
+    "generation_bytes_per_second": 312703.20252531546,
+    "upload_bytes_per_second": 5973164.954195643,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.96530404684428,
+    "estimated_full_shards_per_hour": 67.09882790393446,
     "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": 1.0450683730014134,
-      "bytes": 2703655,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9104454030020861,
       "bytes": 6024195,
@@ -562,6 +556,12 @@
       "bytes": 5630242,
       "pages": 21,
       "shards": 2
+    },
+    {
+      "seconds": 1.3745940619992325,
+      "bytes": 6263311,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 111400, "loss": 0.6208535227924585, "lr": 0.0003, "grad_norm": 0.142977013173407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16301263094, "tokens_per_second": 150214.72800092027, "data_wait_s": 1.4746001397725195e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 111500, "loss": 0.6298375278711319, "lr": 0.0003, "grad_norm": 0.1650104801879643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16315974137, "tokens_per_second": 128759.55248611214, "data_wait_s": 1.6564998077228665e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 111600, "loss": 0.6359089203178883, "lr": 0.0003, "grad_norm": 0.12527254653037218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16330553847, "tokens_per_second": 148136.05730953, "data_wait_s": 1.4473000192083418e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 111700, "loss": 0.5992363654077053, "lr": 0.0003, "grad_norm": 0.13714246636962182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16345018972, "tokens_per_second": 147390.0481111329, "data_wait_s": 1.2933000107295811e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eedb622c368159741e2ecbaf42e4638ca41ea02fef33fd701d49e062838b33ff
-size 567468
+oid sha256:afa165b9adf973b695b343d7f3565b0ab7c4f8150b85171fe1983282fcc46424
+size 569586