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": 74805.39933802045,
-      "required_pages_per_second_upper_estimate": 0.006483942422068329
+      "required_token_positions_per_second": 75137.36532148776,
+      "required_pages_per_second_upper_estimate": 0.0065127163921551825
     },
     {
       "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": 9350.674917252556,
-      "required_pages_per_second_upper_estimate": 0.0037301240295406716
+      "required_token_positions_per_second": 9392.17066518597,
+      "required_pages_per_second_upper_estimate": 0.00374667730380803
     },
     {
       "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": 18701.349834505112,
-      "required_pages_per_second_upper_estimate": 0.004326228151696942
+      "required_token_positions_per_second": 18784.34133037194,
+      "required_pages_per_second_upper_estimate": 0.00434542677366529
     },
     {
       "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": 28052.024751757668,
-      "required_pages_per_second_upper_estimate": 0.0017082211297826538
+      "required_token_positions_per_second": 28176.51199555791,
+      "required_pages_per_second_upper_estimate": 0.001715801749795997
     },
     {
       "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": 28052.024751757668,
-      "required_pages_per_second_upper_estimate": 0.0021595225183648294
+      "required_token_positions_per_second": 28176.51199555791,
+      "required_pages_per_second_upper_estimate": 0.0021691058909953185
     },
     {
       "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": 28052.024751757668,
-      "required_pages_per_second_upper_estimate": 0.34831226332937243
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28176.51199555791,
+      "required_pages_per_second_upper_estimate": 0.3498579782653676
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3162211251797231,
-    "producer_p95_seconds": 0.2998101889988902,
-    "gpu_update_seconds": 0.985597305173761,
-    "producer_batches_per_second": 3.162344069934144,
-    "gpu_batches_per_second": 1.0146131637643834,
-    "producer_headroom": 3.1167977933593156,
+    "producer_batch_seconds": 0.09258320888292815,
+    "producer_p95_seconds": 0.23866539299888245,
+    "gpu_update_seconds": 0.981242816866714,
+    "producer_batches_per_second": 10.801094626829192,
+    "gpu_batches_per_second": 1.0191157405800748,
+    "producer_headroom": 10.598496516873807,
     "sustainable": true,
-    "buffer_seconds": 60.12143561559942,
+    "buffer_seconds": 60.837054645736266,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430180.1909169438,
+    "generation_bytes_per_second": 427837.5590057212,
     "upload_bytes_per_second": 5205976.992258369,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.30665488845095,
+    "estimated_full_shards_per_hour": 91.80398061398246,
     "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": [

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -146,3 +146,4 @@
 {"step": 169600, "loss": 0.5995315574109554, "lr": 0.0003, "grad_norm": 0.1519435142341528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24821573889, "tokens_per_second": 149355.70367058695, "data_wait_s": 1.0183999620494433e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 169700, "loss": 0.55619752779603, "lr": 0.0003, "grad_norm": 0.15390745482804757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24836272163, "tokens_per_second": 149863.83213670508, "data_wait_s": 1.2968999726581387e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 169800, "loss": 0.5832123067229986, "lr": 0.0003, "grad_norm": 0.14845787700264634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24850943091, "tokens_per_second": 150583.05351849974, "data_wait_s": 1.3396000213106163e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 169900, "loss": 0.5867737848311663, "lr": 0.0003, "grad_norm": 0.13999979719676398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24865609258, "tokens_per_second": 149161.79099476011, "data_wait_s": 1.3983999451738782e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a5de01a30db69022dfd242f4fd0931402ce15663cd1b20d26e0d776d257129fc
-size 315195
+oid sha256:a11571dd408ee0adf69dd5d01981c623401164994edfa9f4314f3c26449d32f6
+size 317313