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": 73593.43737921523,
-      "required_pages_per_second_upper_estimate": 0.006378892631168595
+      "required_token_positions_per_second": 74345.90455920962,
+      "required_pages_per_second_upper_estimate": 0.006444114579219877
     },
     {
       "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": 9199.179672401904,
-      "required_pages_per_second_upper_estimate": 0.003669690311313988
+      "required_token_positions_per_second": 9293.238069901203,
+      "required_pages_per_second_upper_estimate": 0.0037072116123748216
     },
     {
       "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": 18398.35934480381,
-      "required_pages_per_second_upper_estimate": 0.004256136634355038
+      "required_token_positions_per_second": 18586.476139802406,
+      "required_pages_per_second_upper_estimate": 0.0042996541441354475
     },
     {
       "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": 27597.539017205712,
-      "required_pages_per_second_upper_estimate": 0.0016805453330508047
+      "required_token_positions_per_second": 27879.71420970361,
+      "required_pages_per_second_upper_estimate": 0.0016977283218151075
     },
     {
       "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": 27597.539017205712,
-      "required_pages_per_second_upper_estimate": 0.002124534948421985
+      "required_token_positions_per_second": 27879.71420970361,
+      "required_pages_per_second_upper_estimate": 0.0021462575758513995
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27597.539017205712,
-      "required_pages_per_second_upper_estimate": 0.3426690715721434
+      "required_token_positions_per_second": 27879.71420970361,
+      "required_pages_per_second_upper_estimate": 0.3461727430833481
     }
   ],
   "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": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.4332356422888779,
-    "producer_p95_seconds": 0.3923835739988135,
-    "gpu_update_seconds": 1.0018284595145541,
-    "producer_batches_per_second": 2.3082126731696935,
-    "gpu_batches_per_second": 0.998174877647776,
-    "producer_headroom": 2.3124331465935652,
+    "producer_batch_seconds": 0.10133354415626172,
+    "producer_p95_seconds": 0.39259150900034,
+    "gpu_update_seconds": 0.9916887882005992,
+    "producer_batches_per_second": 9.868400521528653,
+    "gpu_batches_per_second": 1.008380866959766,
+    "producer_headroom": 9.786382154672912,
     "sustainable": true,
-    "buffer_seconds": 60.10970757087325,
+    "buffer_seconds": 60.49301608023655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385884.31551473273,
-    "upload_bytes_per_second": 5744653.893578308,
+    "generation_bytes_per_second": 385774.034546214,
+    "upload_bytes_per_second": 5793010.327452684,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.80179118234145,
+    "estimated_full_shards_per_hour": 82.77812745370689,
     "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": 0.9770713670004625,
-      "bytes": 6199366,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0861811289942125,
       "bytes": 7695547,
@@ -562,6 +556,12 @@
       "bytes": 740338,
       "pages": 24,
       "shards": 1
+    },
+    {
+      "seconds": 1.4493056469946168,
+      "bytes": 12401988,
+      "pages": 45,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -418,3 +418,4 @@
 {"step": 396800, "loss": 0.600704200565815, "lr": 0.0003, "grad_norm": 0.156986707827914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58080489915, "tokens_per_second": 150732.682964153, "data_wait_s": 1.0575000487733632e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 396900, "loss": 0.6337204575538635, "lr": 0.0003, "grad_norm": 0.18691553438202319, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58095189935, "tokens_per_second": 139834.9905224222, "data_wait_s": 1.1552001524250954e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 397000, "loss": 0.608435932546854, "lr": 0.0003, "grad_norm": 0.16980641858513487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58109934014, "tokens_per_second": 149559.90727095923, "data_wait_s": 1.5167002857197076e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 397100, "loss": 0.5530784148722887, "lr": 0.0003, "grad_norm": 0.142928771927252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58124627355, "tokens_per_second": 149304.01120057673, "data_wait_s": 6.844995368737727e-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 sha256:354d962767780a3b48c9f449a034b11de6908cb240419e0cda7a8d3f7c914425
-size 895989
+oid sha256:2601e1fa7de133f28988863e491eb52bd1d2d0fc4386f8ac6bb8cb31ed08b701
+size 898107