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": 75630.34631104332,
-      "required_pages_per_second_upper_estimate": 0.006555446734880965
+      "required_token_positions_per_second": 73913.29132246837,
+      "required_pages_per_second_upper_estimate": 0.0064066167603073695
     },
     {
       "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": 9453.793288880415,
-      "required_pages_per_second_upper_estimate": 0.003771259489740073
+      "required_token_positions_per_second": 9239.161415308547,
+      "required_pages_per_second_upper_estimate": 0.0036856396263397744
     },
     {
       "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": 18907.58657776083,
-      "required_pages_per_second_upper_estimate": 0.004373937392606438
+      "required_token_positions_per_second": 18478.322830617093,
+      "required_pages_per_second_upper_estimate": 0.004274634779488657
     },
     {
       "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": 28361.379866641248,
-      "required_pages_per_second_upper_estimate": 0.0017270592332182127
+      "required_token_positions_per_second": 27717.48424592564,
+      "required_pages_per_second_upper_estimate": 0.0016878493681758684
     },
     {
       "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": 28361.379866641248,
-      "required_pages_per_second_upper_estimate": 0.0021833375314583344
+      "required_token_positions_per_second": 27717.48424592564,
+      "required_pages_per_second_upper_estimate": 0.002133768664158489
     },
     {
       "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": 28361.379866641248,
-      "required_pages_per_second_upper_estimate": 0.35215341851125875
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27717.48424592564,
+      "required_pages_per_second_upper_estimate": 0.34415838988198766
     }
   ],
   "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": 0.37776527463313414,
-    "producer_p95_seconds": 0.39510837399575394,
-    "gpu_update_seconds": 0.9748467856643207,
-    "producer_batches_per_second": 2.6471464349685068,
-    "gpu_batches_per_second": 1.025802223185809,
-    "producer_headroom": 2.5805621933118146,
+    "producer_batch_seconds": 0.4006766042501795,
+    "producer_p95_seconds": 0.3965510100024403,
+    "gpu_update_seconds": 0.9974931258079148,
+    "producer_batches_per_second": 2.4957783643778897,
+    "gpu_batches_per_second": 1.0025131744041391,
+    "producer_headroom": 2.489521762007066,
     "sustainable": true,
-    "buffer_seconds": 60.44050071118788,
+    "buffer_seconds": 60.84708067428281,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379939.7479929313,
-    "upload_bytes_per_second": 5749751.199731236,
+    "generation_bytes_per_second": 348860.6910331774,
+    "upload_bytes_per_second": 5817052.546362575,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.52622537461237,
+    "estimated_full_shards_per_hour": 74.85738323446743,
     "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": 1.1617618540003605,
-      "bytes": 5816168,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0311880699991889,
       "bytes": 6926958,
@@ -562,6 +556,12 @@
       "bytes": 8174843,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1518778129975544,
+      "bytes": 10667449,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -369,3 +369,4 @@
 {"step": 231900, "loss": 0.6245825625956059, "lr": 0.0003, "grad_norm": 0.17319920183125675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33944115985, "tokens_per_second": 151101.81390666304, "data_wait_s": 1.004899968393147e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 232000, "loss": 0.6067023389041424, "lr": 0.0003, "grad_norm": 0.1522415591060405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33958678080, "tokens_per_second": 148290.96934798567, "data_wait_s": 1.2838994734920561e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 232100, "loss": 0.5992082916200161, "lr": 0.0003, "grad_norm": 0.1648074949305529, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33973124088, "tokens_per_second": 146084.34500625602, "data_wait_s": 1.507100387243554e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 232200, "loss": 0.5945616010576487, "lr": 0.0003, "grad_norm": 0.1442751482313228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33987578625, "tokens_per_second": 146378.1621474991, "data_wait_s": 1.733299723127857e-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 sha256:74b90116606ef8277d18871fcbcd2bc857bb04d80aa2b6b8b3fa77afba24cfbe
-size 791424
+oid sha256:ba154a3ac16db371583a4d084cb90b0d46013ae28debc61f82ac86c8a7b278fb
+size 793542