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": 75534.85615351908,
-      "required_pages_per_second_upper_estimate": 0.0065471698900443544
+      "required_token_positions_per_second": 74430.55937172555,
+      "required_pages_per_second_upper_estimate": 0.006451452243813097
     },
     {
       "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": 9441.857019189885,
-      "required_pages_per_second_upper_estimate": 0.0037664979332973853
+      "required_token_positions_per_second": 9303.819921465694,
+      "required_pages_per_second_upper_estimate": 0.003711432871176677
     },
     {
       "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": 18883.71403837977,
-      "required_pages_per_second_upper_estimate": 0.004368414900762974
+      "required_token_positions_per_second": 18607.63984293139,
+      "required_pages_per_second_upper_estimate": 0.004304549994385884
     },
     {
       "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": 28325.571057569658,
-      "required_pages_per_second_upper_estimate": 0.0017248786646200543
+      "required_token_positions_per_second": 27911.459764397085,
+      "required_pages_per_second_upper_estimate": 0.001699661459539887
     },
     {
       "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": 28325.571057569658,
-      "required_pages_per_second_upper_estimate": 0.0021805808702108757
+      "required_token_positions_per_second": 27911.459764397085,
+      "required_pages_per_second_upper_estimate": 0.002148701436529023
     },
     {
       "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": 28325.571057569658,
-      "required_pages_per_second_upper_estimate": 0.3517087929469642
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27911.459764397085,
+      "required_pages_per_second_upper_estimate": 0.34656691662710415
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.10160118406261631,
-    "producer_p95_seconds": 0.31357375999868964,
-    "gpu_update_seconds": 0.9760791739664297,
-    "producer_batches_per_second": 9.842404980080792,
-    "gpu_batches_per_second": 1.0245070550336248,
-    "producer_headroom": 9.606966522800334,
+    "producer_batch_seconds": 0.1095162872815365,
+    "producer_p95_seconds": 0.3171666960006405,
+    "gpu_update_seconds": 0.9905608747582186,
+    "producer_batches_per_second": 9.131061916199485,
+    "gpu_batches_per_second": 1.0095290713395937,
+    "producer_headroom": 9.044872679182017,
     "sustainable": true,
-    "buffer_seconds": 60.516908785918645,
+    "buffer_seconds": 60.424213360251336,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356980.45639034174,
-    "upload_bytes_per_second": 4202883.174415789,
+    "generation_bytes_per_second": 381697.77455599024,
+    "upload_bytes_per_second": 4100456.994232863,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.59969586165131,
+    "estimated_full_shards_per_hour": 81.90345695028095,
     "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": 1.4371028810019197,
-      "bytes": 13333064,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 0.925103517001844,
       "bytes": 1157884,
@@ -562,6 +556,12 @@
       "bytes": 7478866,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9801313110001502,
+      "bytes": 949416,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -283,3 +283,4 @@
 {"step": 343300, "loss": 0.5748323947191238, "lr": 0.0003, "grad_norm": 0.17213277630921067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50249887136, "tokens_per_second": 148665.9782444599, "data_wait_s": 1.631099803489633e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 343400, "loss": 0.5832381173968315, "lr": 0.0003, "grad_norm": 0.1490679566560398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50264490100, "tokens_per_second": 149142.0098490905, "data_wait_s": 1.70599996636156e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 343500, "loss": 0.6186619400978088, "lr": 0.0003, "grad_norm": 0.17305364622911118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50279126614, "tokens_per_second": 148423.57161131405, "data_wait_s": 1.7221998859895393e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 343600, "loss": 0.546442705206573, "lr": 0.0003, "grad_norm": 0.15118177201015093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50293865040, "tokens_per_second": 150371.70046929838, "data_wait_s": 1.4756999007659033e-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:b47e07e84bdbad272391af7fb7945c84604086e5637561aa473c6bc340f7a225
-size 607710
+oid sha256:d431f747f47bda5309296be29e3c71c77b8b708fa9a15b27997345215eca8be9
+size 609828