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": 75922.91687064731,
-      "required_pages_per_second_upper_estimate": 0.006580806009474134
+      "required_token_positions_per_second": 74627.94042569294,
+      "required_pages_per_second_upper_estimate": 0.006468560733313282
     },
     {
       "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": 9490.364608830914,
-      "required_pages_per_second_upper_estimate": 0.003785848336058287
+      "required_token_positions_per_second": 9328.492553211618,
+      "required_pages_per_second_upper_estimate": 0.003721275152868844
     },
     {
       "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": 18980.72921766183,
-      "required_pages_per_second_upper_estimate": 0.004390857655081038
+      "required_token_positions_per_second": 18656.985106423235,
+      "required_pages_per_second_upper_estimate": 0.004315965152647747
     },
     {
       "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": 28471.093826492743,
-      "required_pages_per_second_upper_estimate": 0.001733740237748508
+      "required_token_positions_per_second": 27985.477659634853,
+      "required_pages_per_second_upper_estimate": 0.0017041687610179833
     },
     {
       "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": 28471.093826492743,
-      "required_pages_per_second_upper_estimate": 0.0021917836157953814
+      "required_token_positions_per_second": 27985.477659634853,
+      "required_pages_per_second_upper_estimate": 0.002154399538999074
     },
     {
       "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": 28471.093826492743,
-      "required_pages_per_second_upper_estimate": 0.35351569870361127
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27985.477659634853,
+      "required_pages_per_second_upper_estimate": 0.3474859711639973
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.07895137357800763,
-    "producer_p95_seconds": 0.30766253099864116,
-    "gpu_update_seconds": 0.9710901930389888,
-    "producer_batches_per_second": 12.666024094083093,
-    "gpu_batches_per_second": 1.0297704653679378,
-    "producer_headroom": 12.299851782559635,
+    "producer_batch_seconds": 0.2566958453592463,
+    "producer_p95_seconds": 0.3140008519985713,
+    "gpu_update_seconds": 0.9879409719662703,
+    "producer_batches_per_second": 3.89566102482297,
+    "gpu_batches_per_second": 1.0122062232217468,
+    "producer_headroom": 3.848683139314722,
     "sustainable": true,
-    "buffer_seconds": 60.207591968417304,
+    "buffer_seconds": 60.26439928994249,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 172,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 584382.1328700304,
-    "upload_bytes_per_second": 4949056.557872818,
+    "generation_bytes_per_second": 545638.3041002426,
+    "upload_bytes_per_second": 4996239.710866044,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.69740099704591,
+    "estimated_full_shards_per_hour": 117.08127825026949,
     "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": 1.0801306080029462,
-      "bytes": 6178099,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0223706500000844,
       "bytes": 2585543,
@@ -562,6 +556,12 @@
       "bytes": 2351070,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2001916299996083,
+      "bytes": 10729681,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -283,3 +283,4 @@
 {"step": 383300, "loss": 0.59818534553051, "lr": 0.0003, "grad_norm": 0.18368050872323263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56104023859, "tokens_per_second": 149688.41897770754, "data_wait_s": 1.0154002666240558e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 383400, "loss": 0.6379445008933544, "lr": 0.0003, "grad_norm": 0.15359044005996228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56118712420, "tokens_per_second": 150167.46518709254, "data_wait_s": 1.058599809766747e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 383500, "loss": 0.5937663409858942, "lr": 0.0003, "grad_norm": 0.15485348331098175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56133445333, "tokens_per_second": 149660.40936670825, "data_wait_s": 9.308001608587801e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 383600, "loss": 0.5732032470405102, "lr": 0.0003, "grad_norm": 0.14635394214529898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56148144294, "tokens_per_second": 149594.13238662592, "data_wait_s": 9.798000974114984e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae634070813343bae686dd1e69d3dacaaf66bd83a0563b374a0950170d5f0817
-size 607710
+oid sha256:75bef2d78d003ed91064049933ef82e2481a3ec9723e1714aba774db8980fa73
+size 609828