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": 74293.0989019505,
-      "required_pages_per_second_upper_estimate": 0.006439537518683372
+      "required_token_positions_per_second": 75682.13299016481,
+      "required_pages_per_second_upper_estimate": 0.006559935472975077
     },
     {
       "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": 9286.637362743813,
-      "required_pages_per_second_upper_estimate": 0.0037045784916003724
+      "required_token_positions_per_second": 9460.266623770602,
+      "required_pages_per_second_upper_estimate": 0.00377384179981275
     },
     {
       "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": 18573.274725487627,
-      "required_pages_per_second_upper_estimate": 0.00429660022927068
+      "required_token_positions_per_second": 18920.533247541203,
+      "required_pages_per_second_upper_estimate": 0.004376932376806522
     },
     {
       "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": 27859.912088231442,
-      "required_pages_per_second_upper_estimate": 0.0016965224765111545
+      "required_token_positions_per_second": 28380.799871311803,
+      "required_pages_per_second_upper_estimate": 0.0017282418096137594
     },
     {
       "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": 27859.912088231442,
-      "required_pages_per_second_upper_estimate": 0.002144733153724691
+      "required_token_positions_per_second": 28380.799871311803,
+      "required_pages_per_second_upper_estimate": 0.0021848325371758874
     },
     {
       "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": 27859.912088231442,
-      "required_pages_per_second_upper_estimate": 0.34592686700810116
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28380.799871311803,
+      "required_pages_per_second_upper_estimate": 0.35239454997469244
     }
   ],
   "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": 0.4520289101327819,
-    "producer_p95_seconds": 0.3542765300007886,
-    "gpu_update_seconds": 0.9923936555305587,
-    "producer_batches_per_second": 2.2122478841148756,
-    "gpu_batches_per_second": 1.007664644394945,
-    "producer_headroom": 2.195420764656505,
+    "producer_batch_seconds": 0.08517284599986397,
+    "producer_p95_seconds": 0.3263877190001949,
+    "gpu_update_seconds": 0.9741797315567366,
+    "producer_batches_per_second": 11.740831109501697,
+    "gpu_batches_per_second": 1.0265046249751086,
+    "producer_headroom": 11.437679698507345,
     "sustainable": true,
-    "buffer_seconds": 60.53601298736408,
+    "buffer_seconds": 60.39914335651767,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332550.12303666444,
-    "upload_bytes_per_second": 5076615.40048203,
+    "generation_bytes_per_second": 331535.26427497296,
+    "upload_bytes_per_second": 5059672.6514437925,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.3575150330062,
+    "estimated_full_shards_per_hour": 71.13974996745006,
     "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": 0.9833170429992606,
-      "bytes": 2688996,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4986307230028615,
       "bytes": 10640144,
@@ -562,6 +556,12 @@
       "bytes": 11156038,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 0.9720597020022979,
+      "bytes": 1226766,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -319,3 +319,4 @@
 {"step": 386900, "loss": 0.6079707313328981, "lr": 0.0003, "grad_norm": 0.15555526425994035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56631539131, "tokens_per_second": 148225.39397405653, "data_wait_s": 1.2568998499773443e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 387000, "loss": 0.5514918360859156, "lr": 0.0003, "grad_norm": 0.1510423615634618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56646036656, "tokens_per_second": 146850.150214591, "data_wait_s": 9.527997463010252e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 387100, "loss": 0.5839230827987194, "lr": 0.0003, "grad_norm": 0.17005130184964223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56660762388, "tokens_per_second": 150913.89009359706, "data_wait_s": 1.0976000339724123e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 387200, "loss": 0.5735466405749321, "lr": 0.0003, "grad_norm": 0.14504306351229077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56675586356, "tokens_per_second": 150590.1287857146, "data_wait_s": 1.0994001058861613e-05, "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 sha256:9fb454553889d9885fb9565af0a6187451c8780579fdde5b9fc2d54aa2042938
-size 684741
+oid sha256:deb8ee7dd323ab86615922fc9e81a7e02dcca87df67f635c29c41e39ffb64bf0
+size 686859