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": 74115.57923270097,
-      "required_pages_per_second_upper_estimate": 0.00642415056367231
+      "required_token_positions_per_second": 75560.3025058102,
+      "required_pages_per_second_upper_estimate": 0.006549375515367759
     },
     {
       "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": 9264.447404087621,
-      "required_pages_per_second_upper_estimate": 0.00369572658532297
+      "required_token_positions_per_second": 9445.037813226276,
+      "required_pages_per_second_upper_estimate": 0.0037677667995956103
     },
     {
       "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": 18528.894808175242,
-      "required_pages_per_second_upper_estimate": 0.004286333716460328
+      "required_token_positions_per_second": 18890.07562645255,
+      "required_pages_per_second_upper_estimate": 0.004369886542203485
     },
     {
       "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": 27793.342212262865,
-      "required_pages_per_second_upper_estimate": 0.0016924687203298148
+      "required_token_positions_per_second": 28335.11343967883,
+      "required_pages_per_second_upper_estimate": 0.0017254597456254931
     },
     {
       "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": 27793.342212262865,
-      "required_pages_per_second_upper_estimate": 0.0021396084203954193
+      "required_token_positions_per_second": 28335.11343967883,
+      "required_pages_per_second_upper_estimate": 0.002181315469200644
     },
     {
       "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": 27793.342212262865,
-      "required_pages_per_second_upper_estimate": 0.3451002919436143
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28335.11343967883,
+      "required_pages_per_second_upper_estimate": 0.35182727739646164
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.40174102159386393,
-    "producer_p95_seconds": 0.29589034200034803,
-    "gpu_update_seconds": 0.9947706104881932,
-    "producer_batches_per_second": 2.4891657716023334,
-    "gpu_batches_per_second": 1.005256879783813,
-    "producer_headroom": 2.4761489542231674,
+    "producer_batch_seconds": 0.12938273214066953,
+    "producer_p95_seconds": 0.3352755679989059,
+    "gpu_update_seconds": 0.975750460955747,
+    "producer_batches_per_second": 7.7290066723337105,
+    "gpu_batches_per_second": 1.0248521932754207,
+    "producer_headroom": 7.541581823259662,
     "sustainable": true,
-    "buffer_seconds": 60.68100723977978,
+    "buffer_seconds": 60.496528579256314,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337462.42987067625,
-    "upload_bytes_per_second": 5405104.392031946,
+    "generation_bytes_per_second": 340172.2969059776,
+    "upload_bytes_per_second": 5446769.534795594,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.41158172693459,
+    "estimated_full_shards_per_hour": 72.99305611023422,
     "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": 1.0015506020008615,
-      "bytes": 7818449,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8705007850003312,
       "bytes": 1070987,
@@ -562,6 +556,12 @@
       "bytes": 6352582,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1081351329994504,
+      "bytes": 11588439,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -217,3 +217,4 @@
 {"step": 336700, "loss": 0.5935948863625526, "lr": 0.0003, "grad_norm": 0.14299183902306487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49284108645, "tokens_per_second": 148462.8509218014, "data_wait_s": 1.3539996871259063e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 336800, "loss": 0.5988835487514734, "lr": 0.0003, "grad_norm": 0.15996886367219212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49298928501, "tokens_per_second": 140223.44892660063, "data_wait_s": 9.431998478248715e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 336900, "loss": 0.6639048252254725, "lr": 0.0003, "grad_norm": 0.16959647781398787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49313718546, "tokens_per_second": 149526.03471198716, "data_wait_s": 1.549800072098151e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 337000, "loss": 0.5864039976149797, "lr": 0.0003, "grad_norm": 0.17749098000646407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49328423295, "tokens_per_second": 148740.57350397322, "data_wait_s": 1.7457001376897097e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f47752702914c9f671677e74f0d4106fbab32da552b0acbfdae1296f95464e14
-size 467139
+oid sha256:33f3b6acfccc23933edf2184d941f05157d3a83b131322f8bf6b1e30610b99dd
+size 469257