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": 74425.37698485564,
-      "required_pages_per_second_upper_estimate": 0.0064510030476538525
+      "required_token_positions_per_second": 75739.99025522982,
+      "required_pages_per_second_upper_estimate": 0.006564950394073014
     },
     {
       "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": 9303.172123106955,
-      "required_pages_per_second_upper_estimate": 0.0037111744547259273
+      "required_token_positions_per_second": 9467.498781903727,
+      "required_pages_per_second_upper_estimate": 0.0037767268158224536
     },
     {
       "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": 18606.34424621391,
-      "required_pages_per_second_upper_estimate": 0.004304250280887018
+      "required_token_positions_per_second": 18934.997563807454,
+      "required_pages_per_second_upper_estimate": 0.004380278441811447
     },
     {
       "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": 27909.516369320863,
-      "required_pages_per_second_upper_estimate": 0.0016995431169759508
+      "required_token_positions_per_second": 28402.49634571118,
+      "required_pages_per_second_upper_estimate": 0.0017295630110720777
     },
     {
       "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": 27909.516369320863,
-      "required_pages_per_second_upper_estimate": 0.0021485518285963936
+      "required_token_positions_per_second": 28402.49634571118,
+      "required_pages_per_second_upper_estimate": 0.0021865027918348367
     },
     {
       "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": 27909.516369320863,
-      "required_pages_per_second_upper_estimate": 0.3465427861643824
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28402.49634571118,
+      "required_pages_per_second_upper_estimate": 0.35266394757330394
     }
   ],
   "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": 0.15256538671872022,
-    "producer_p95_seconds": 0.3935451830038801,
-    "gpu_update_seconds": 0.9906298494800028,
-    "producer_batches_per_second": 6.554566677982255,
-    "gpu_batches_per_second": 1.0094587807190705,
-    "producer_headroom": 6.493149401616202,
+    "producer_batch_seconds": 0.14914389588273025,
+    "producer_p95_seconds": 0.3874536770017585,
+    "gpu_update_seconds": 0.9734355622644024,
+    "producer_batches_per_second": 6.704934144849521,
+    "gpu_batches_per_second": 1.0272893643558731,
+    "producer_headroom": 6.526821339237384,
     "sustainable": true,
-    "buffer_seconds": 60.428420818280166,
+    "buffer_seconds": 60.35300486039295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300331.81370173325,
-    "upload_bytes_per_second": 5967557.130011284,
+    "generation_bytes_per_second": 300258.5200801012,
+    "upload_bytes_per_second": 5910628.800790494,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.44421585358617,
+    "estimated_full_shards_per_hour": 64.42848874857214,
     "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": 1.19388542699744,
-      "bytes": 12323550,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.0270863580008154,
       "bytes": 6920922,
@@ -562,6 +556,12 @@
       "bytes": 7189534,
       "pages": 45,
       "shards": 2
+    },
+    {
+      "seconds": 1.0300960800013854,
+      "bytes": 7293613,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -336,3 +336,4 @@
 {"step": 118600, "loss": 0.6688226964324713, "lr": 0.0003, "grad_norm": 0.15798738859772068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17355731774, "tokens_per_second": 145073.12589904512, "data_wait_s": 1.6703001165296882e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 118700, "loss": 0.6316186003386974, "lr": 0.0003, "grad_norm": 0.14607130050516742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17370383301, "tokens_per_second": 148738.52458228893, "data_wait_s": 1.0177995136473328e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 118800, "loss": 0.6226529106497765, "lr": 0.0003, "grad_norm": 0.14603244399728735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17384892783, "tokens_per_second": 148250.71523810277, "data_wait_s": 1.9462997443042696e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 118900, "loss": 0.5921273212879896, "lr": 0.0003, "grad_norm": 0.13395452390409357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17399398353, "tokens_per_second": 143621.63730805874, "data_wait_s": 1.3270997442305088e-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:d5a6decc8ef4088581c7f3a2981d7bfdf4e8dbbb8a331e8967973f261e9a72f2
-size 720747
+oid sha256:5047f9e7177725d8e1ad4378abec62a21ac46f050426122d8a3bf6a26f464628
+size 722865