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": 74368.94689165257,
-      "required_pages_per_second_upper_estimate": 0.006446111830195773
+      "required_token_positions_per_second": 75868.04310435671,
+      "required_pages_per_second_upper_estimate": 0.006576049690488354
     },
     {
       "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": 9296.11836145657,
-      "required_pages_per_second_upper_estimate": 0.0037083606037404543
+      "required_token_positions_per_second": 9483.505388044588,
+      "required_pages_per_second_upper_estimate": 0.003783112090332132
     },
     {
       "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": 18592.23672291314,
-      "required_pages_per_second_upper_estimate": 0.004300986753655316
+      "required_token_positions_per_second": 18967.010776089177,
+      "required_pages_per_second_upper_estimate": 0.004387684135059531
     },
     {
       "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": 27888.355084369712,
-      "required_pages_per_second_upper_estimate": 0.0016982545057471076
+      "required_token_positions_per_second": 28450.516164133765,
+      "required_pages_per_second_upper_estimate": 0.001732487166073497
     },
     {
       "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": 27888.355084369712,
-      "required_pages_per_second_upper_estimate": 0.002146922774302668
+      "required_token_positions_per_second": 28450.516164133765,
+      "required_pages_per_second_upper_estimate": 0.0021901994903843723
     },
     {
       "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": 27888.355084369712,
-      "required_pages_per_second_upper_estimate": 0.34628003382755396
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28450.516164133765,
+      "required_pages_per_second_upper_estimate": 0.3532601930061185
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.14866582536740225,
-    "producer_p95_seconds": 0.32602170600148384,
-    "gpu_update_seconds": 0.9913815252408192,
-    "producer_batches_per_second": 6.726495464096543,
-    "gpu_batches_per_second": 1.008693398595548,
-    "producer_headroom": 6.668523332721483,
+    "producer_batch_seconds": 0.12452985720315723,
+    "producer_p95_seconds": 0.32537357800174505,
+    "gpu_update_seconds": 0.971792562233178,
+    "producer_batches_per_second": 8.03020273578734,
+    "gpu_batches_per_second": 1.0290261922791437,
+    "producer_headroom": 7.803691291862655,
     "sustainable": true,
-    "buffer_seconds": 60.474273039689976,
+    "buffer_seconds": 60.25113885845704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332564.5311415766,
-    "upload_bytes_per_second": 4939741.117864166,
+    "generation_bytes_per_second": 333278.1450406388,
+    "upload_bytes_per_second": 4994742.319966547,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.36060667691682,
+    "estimated_full_shards_per_hour": 71.5137316075742,
     "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": 1.3262504000012996,
-      "bytes": 7612878,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.397075862998463,
       "bytes": 11930697,
@@ -562,6 +556,12 @@
       "bytes": 10502441,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.2662762959989777,
+      "bytes": 11950298,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -255,3 +255,4 @@
 {"step": 380500, "loss": 0.5221319552510977, "lr": 0.0003, "grad_norm": 0.1407149522118551, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55693639735, "tokens_per_second": 150086.71387165552, "data_wait_s": 1.2709999282378703e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 380600, "loss": 0.5683339741080999, "lr": 0.0003, "grad_norm": 0.13495020868309707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55708230802, "tokens_per_second": 147894.6098215744, "data_wait_s": 8.877999789547175e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 380700, "loss": 0.6049262899905443, "lr": 0.0003, "grad_norm": 0.1816526551864626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55722882126, "tokens_per_second": 150152.3056610104, "data_wait_s": 8.253002306446433e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 380800, "loss": 0.5672803446650505, "lr": 0.0003, "grad_norm": 0.16266216004883965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55737586324, "tokens_per_second": 149288.4222409075, "data_wait_s": 9.526000212645158e-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 sha256:651ddf91dc256757376ef314863904302861ec61d36eac722c647945f7341179
-size 548406
+oid sha256:283df3d91ae5e61049a01a84de548b567209a9ffa0a4ddbc8ae60c186cc45384
+size 550524