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": 75666.67043597525,
-      "required_pages_per_second_upper_estimate": 0.006558595217967411
+      "required_token_positions_per_second": 74808.50644745442,
+      "required_pages_per_second_upper_estimate": 0.006484211738439169
     },
     {
       "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": 9458.333804496906,
-      "required_pages_per_second_upper_estimate": 0.0037730707693062492
+      "required_token_positions_per_second": 9351.063305931802,
+      "required_pages_per_second_upper_estimate": 0.0037302789635917516
     },
     {
       "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": 18916.667608993812,
-      "required_pages_per_second_upper_estimate": 0.004376038129361498
+      "required_token_positions_per_second": 18702.126611863605,
+      "required_pages_per_second_upper_estimate": 0.004326407845468014
     },
     {
       "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": 28375.001413490718,
-      "required_pages_per_second_upper_estimate": 0.0017278887139546124
+      "required_token_positions_per_second": 28053.189917795407,
+      "required_pages_per_second_upper_estimate": 0.0017082920822811983
     },
     {
       "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": 28375.001413490718,
-      "required_pages_per_second_upper_estimate": 0.0021843861558416616
+      "required_token_positions_per_second": 28053.189917795407,
+      "required_pages_per_second_upper_estimate": 0.002159612216071801
     },
     {
       "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": 28375.001413490718,
-      "required_pages_per_second_upper_estimate": 0.3523225525347445
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28053.189917795407,
+      "required_pages_per_second_upper_estimate": 0.34832673079200127
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12722127366407676,
-    "producer_p95_seconds": 0.4442140789979021,
-    "gpu_update_seconds": 0.9743788060872106,
-    "producer_batches_per_second": 7.8603206146203535,
-    "gpu_batches_per_second": 1.0262949006615567,
-    "producer_headroom": 7.658929815936469,
+    "producer_batch_seconds": 0.19924010114851853,
+    "producer_p95_seconds": 0.43951476000074763,
+    "gpu_update_seconds": 0.9855563692049731,
+    "producer_batches_per_second": 5.019069927366555,
+    "gpu_batches_per_second": 1.0146553066332251,
+    "producer_headroom": 4.94657633440125,
     "sustainable": true,
-    "buffer_seconds": 60.41148597740706,
+    "buffer_seconds": 60.11893852150336,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332267.7093581426,
-    "upload_bytes_per_second": 5677308.77381865,
+    "generation_bytes_per_second": 304391.2570648404,
+    "upload_bytes_per_second": 5680028.704511996,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.29691563184936,
+    "estimated_full_shards_per_hour": 65.31527790030393,
     "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": 1.0235948650006321,
-      "bytes": 9738901,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.8542503440039582,
       "bytes": 2305951,
@@ -562,6 +556,12 @@
       "bytes": 5947609,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3013317000004463,
+      "bytes": 11518633,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -413,3 +413,4 @@
 {"step": 236300, "loss": 0.5963732190430164, "lr": 0.0003, "grad_norm": 0.1587367101003753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34587718002, "tokens_per_second": 148591.4770712124, "data_wait_s": 1.204499858431518e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 236400, "loss": 0.6338792107999325, "lr": 0.0003, "grad_norm": 0.14740172653696962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34602370900, "tokens_per_second": 148298.6080736068, "data_wait_s": 1.0199000826105475e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 236500, "loss": 0.5867517907172441, "lr": 0.0003, "grad_norm": 0.14828568283742657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34616852937, "tokens_per_second": 146583.28181564133, "data_wait_s": 1.745999907143414e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 236600, "loss": 0.5741153005510569, "lr": 0.0003, "grad_norm": 0.16989412711838428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34631263270, "tokens_per_second": 145692.82722657136, "data_wait_s": 1.1872005416080356e-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 sha256:9cec22fd1fa8cdb040ec684fa4f87a0b94ce0f07f8c26cc9801b94bbe69cf58b
-size 885399
+oid sha256:71f78a21faddca59108504688c4b2ee900cdfea9d4de172c905dcddb1b01da4a
+size 887517