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": 76044.87429931147,
-      "required_pages_per_second_upper_estimate": 0.0065913769702924665
+      "required_token_positions_per_second": 74808.33735482603,
+      "required_pages_per_second_upper_estimate": 0.0064841970819187016
     },
     {
       "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": 9505.609287413934,
-      "required_pages_per_second_upper_estimate": 0.0037919296662733103
+      "required_token_positions_per_second": 9351.042169353253,
+      "required_pages_per_second_upper_estimate": 0.0037302705318945483
     },
     {
       "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": 19011.218574827868,
-      "required_pages_per_second_upper_estimate": 0.004397910831267042
+      "required_token_positions_per_second": 18702.084338706507,
+      "required_pages_per_second_upper_estimate": 0.0043263980663171015
     },
     {
       "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": 28516.827862241804,
-      "required_pages_per_second_upper_estimate": 0.0017365252005776845
+      "required_token_positions_per_second": 28053.12650805976,
+      "required_pages_per_second_upper_estimate": 0.0017082882209609824
     },
     {
       "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": 28516.827862241804,
-      "required_pages_per_second_upper_estimate": 0.002195304348467258
+      "required_token_positions_per_second": 28053.12650805976,
+      "required_pages_per_second_upper_estimate": 0.002159607334614821
     },
     {
       "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": 28516.827862241804,
-      "required_pages_per_second_upper_estimate": 0.35408356236564315
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28053.12650805976,
+      "required_pages_per_second_upper_estimate": 0.34832594345530327
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.07886946051560528,
-    "producer_p95_seconds": 0.20136157300112245,
-    "gpu_update_seconds": 0.9695328012484801,
-    "producer_batches_per_second": 12.679178904769328,
-    "gpu_batches_per_second": 1.031424618860019,
-    "producer_headroom": 12.292879841071644,
+    "producer_batch_seconds": 0.27917365273438577,
+    "producer_p95_seconds": 0.20891276099973766,
+    "gpu_update_seconds": 0.9855585969020827,
+    "producer_batches_per_second": 3.5819999136932528,
+    "gpu_batches_per_second": 1.0146530131676708,
+    "producer_headroom": 3.5302708090429036,
     "sustainable": true,
-    "buffer_seconds": 60.111033677405764,
+    "buffer_seconds": 60.11907441102704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441911.2600301277,
-    "upload_bytes_per_second": 5674576.129874926,
+    "generation_bytes_per_second": 378330.52753664996,
+    "upload_bytes_per_second": 5685180.552910053,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.82386923482774,
+    "estimated_full_shards_per_hour": 81.18092412542939,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 7585534,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0756502530002763,
+      "bytes": 6680691,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -82,3 +82,4 @@
 {"step": 363200, "loss": 0.6068327557295561, "lr": 0.0003, "grad_norm": 0.1656343598441726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53162106222, "tokens_per_second": 148960.51610466884, "data_wait_s": 1.5003999578766525e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 363300, "loss": 0.6225243136286736, "lr": 0.0003, "grad_norm": 0.18326852934646368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53176756150, "tokens_per_second": 149920.7606125064, "data_wait_s": 1.2214000889798626e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 363400, "loss": 0.5955623667687178, "lr": 0.0003, "grad_norm": 0.15571062923710338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53191428154, "tokens_per_second": 142148.8241390262, "data_wait_s": 7.74100044509396e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 363500, "loss": 0.6045342851430178, "lr": 0.0003, "grad_norm": 0.1777388046768215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53206038919, "tokens_per_second": 149298.15606967517, "data_wait_s": 1.7934000425157137e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8757d54388d570f6d2391d1009793513b57ad1ccf77acc18202b8b3cbc77f67
-size 178860
+oid sha256:007ae204df85e1904d0b88fd8f593dd73d9cd9b5b111864dc330fb342fe1b12c
+size 180978