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": 74544.97842734645,
-      "required_pages_per_second_upper_estimate": 0.006461369797561871
+      "required_token_positions_per_second": 75335.94636392311,
+      "required_pages_per_second_upper_estimate": 0.006529928893614429
     },
     {
       "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": 9318.122303418306,
-      "required_pages_per_second_upper_estimate": 0.0037171383051772403
+      "required_token_positions_per_second": 9416.993295490389,
+      "required_pages_per_second_upper_estimate": 0.003756579422167859
     },
     {
       "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": 18636.24460683661,
-      "required_pages_per_second_upper_estimate": 0.004311167203088707
+      "required_token_positions_per_second": 18833.986590980778,
+      "required_pages_per_second_upper_estimate": 0.0043569113309804
     },
     {
       "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": 27954.36691025492,
-      "required_pages_per_second_upper_estimate": 0.0017022742796062326
+      "required_token_positions_per_second": 28250.979886471167,
+      "required_pages_per_second_upper_estimate": 0.001720336453649786
     },
     {
       "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": 27954.36691025492,
-      "required_pages_per_second_upper_estimate": 0.0021520045473917408
+      "required_token_positions_per_second": 28250.979886471167,
+      "required_pages_per_second_upper_estimate": 0.002174838635378137
     },
     {
       "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": 27954.36691025492,
-      "required_pages_per_second_upper_estimate": 0.34709967977767886
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28250.979886471167,
+      "required_pages_per_second_upper_estimate": 0.3507826202425117
     }
   ],
   "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": 0.12311220349948826,
-    "producer_p95_seconds": 0.4010945720001473,
-    "gpu_update_seconds": 0.9890404632936785,
-    "producer_batches_per_second": 8.12267160829557,
-    "gpu_batches_per_second": 1.0110809791035489,
-    "producer_headroom": 8.03365089065106,
+    "producer_batch_seconds": 0.16328131734337603,
+    "producer_p95_seconds": 0.4133406179971644,
+    "gpu_update_seconds": 0.978656319572125,
+    "producer_batches_per_second": 6.124399387941169,
+    "gpu_batches_per_second": 1.0218091683474815,
+    "producer_headroom": 5.993682164592279,
     "sustainable": true,
-    "buffer_seconds": 60.331468260914384,
+    "buffer_seconds": 60.67669181347175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316222.72737043106,
-    "upload_bytes_per_second": 5948432.450447036,
+    "generation_bytes_per_second": 326728.9085583137,
+    "upload_bytes_per_second": 5974885.961383253,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.85403600535106,
+    "estimated_full_shards_per_hour": 70.10841791688974,
     "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": 1.1510760689998278,
-      "bytes": 5630242,
-      "pages": 21,
-      "shards": 2
-    },
     {
       "seconds": 1.3745940619992325,
       "bytes": 6263311,
@@ -562,6 +556,12 @@
       "bytes": 6096964,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0299214569968171,
+      "bytes": 6792847,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -341,3 +341,4 @@
 {"step": 119100, "loss": 0.5712260492146015, "lr": 0.0003, "grad_norm": 0.13741100760686165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17428635994, "tokens_per_second": 148904.5290307736, "data_wait_s": 1.4810997527092695e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 119200, "loss": 0.5859232051298022, "lr": 0.0003, "grad_norm": 0.13938095464892203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17443251462, "tokens_per_second": 148488.68158974353, "data_wait_s": 1.6275000234600157e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 119300, "loss": 0.6423541735857725, "lr": 0.0003, "grad_norm": 0.13440198642835463, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17457970719, "tokens_per_second": 149450.31556604808, "data_wait_s": 1.8403996364213526e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 119400, "loss": 0.6104825809597969, "lr": 0.0003, "grad_norm": 0.16794767334880456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17472747252, "tokens_per_second": 149110.36216746346, "data_wait_s": 1.1146999895572662e-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:a1dc3e5607dc3d329c250208e21a4006fc2ac9884865e8fbfc065a36124713af
-size 731337
+oid sha256:444363139c9ac6cb291f0eb1dad205bd9443c10db98177c266ae43b1718d3317
+size 733455