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": 74874.00403465243,
-      "required_pages_per_second_upper_estimate": 0.006489888903295376
+      "required_token_positions_per_second": 75626.8274640585,
+      "required_pages_per_second_upper_estimate": 0.006555141730142756
     },
     {
       "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": 9359.250504331554,
-      "required_pages_per_second_upper_estimate": 0.0037335449594429367
+      "required_token_positions_per_second": 9453.353433007313,
+      "required_pages_per_second_upper_estimate": 0.003771084024655861
     },
     {
       "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": 18718.50100866311,
-      "required_pages_per_second_upper_estimate": 0.004330195773988038
+      "required_token_positions_per_second": 18906.706866014625,
+      "required_pages_per_second_upper_estimate": 0.0043737338867763514
     },
     {
       "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": 28077.751512994662,
-      "required_pages_per_second_upper_estimate": 0.0017097877545641021
+      "required_token_positions_per_second": 28360.060299021938,
+      "required_pages_per_second_upper_estimate": 0.001726978878473433
     },
     {
       "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": 28077.751512994662,
-      "required_pages_per_second_upper_estimate": 0.0021615030356610864
+      "required_token_positions_per_second": 28360.060299021938,
+      "required_pages_per_second_upper_estimate": 0.002183235947490203
     },
     {
       "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": 28077.751512994662,
-      "required_pages_per_second_upper_estimate": 0.3486317036020048
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28360.060299021938,
+      "required_pages_per_second_upper_estimate": 0.35213703389773565
     }
   ],
   "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": 0.15375935694578402,
-    "producer_p95_seconds": 0.3801806269984809,
-    "gpu_update_seconds": 0.9846942333400248,
-    "producer_batches_per_second": 6.503669239151428,
-    "gpu_batches_per_second": 1.0155436745151425,
-    "producer_headroom": 6.404125595343318,
+    "producer_batch_seconds": 0.16053574478928567,
+    "producer_p95_seconds": 0.3767570310010342,
+    "gpu_update_seconds": 0.9748921443920028,
+    "producer_batches_per_second": 6.229142309163417,
+    "gpu_batches_per_second": 1.0257544957690226,
+    "producer_headroom": 6.0727419035032755,
     "sustainable": true,
-    "buffer_seconds": 60.066348233741515,
+    "buffer_seconds": 60.443312952304176,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 273630.02016605664,
-    "upload_bytes_per_second": 5233898.685066041,
+    "generation_bytes_per_second": 268873.1664870888,
+    "upload_bytes_per_second": 5205347.348566171,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.71463254677081,
+    "estimated_full_shards_per_hour": 57.6939224811506,
     "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": 1.1656662089990277,
-      "bytes": 11880694,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.2683707119977043,
       "bytes": 2325676,
@@ -562,6 +556,12 @@
       "bytes": 8176480,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.249701370994444,
+      "bytes": 10043413,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -353,3 +353,4 @@
 {"step": 390300, "loss": 0.6017175558954477, "lr": 0.0003, "grad_norm": 0.1479221209816008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57129537823, "tokens_per_second": 151049.98714732862, "data_wait_s": 8.486000297125429e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 390400, "loss": 0.6008829958736897, "lr": 0.0003, "grad_norm": 0.14908405713365075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57144158581, "tokens_per_second": 142276.7659792644, "data_wait_s": 1.5287005226127803e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 390500, "loss": 0.559561824426055, "lr": 0.0003, "grad_norm": 0.1650607503399263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57158681334, "tokens_per_second": 148375.22940628073, "data_wait_s": 9.72800626186654e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 390600, "loss": 0.5853303074836731, "lr": 0.0003, "grad_norm": 0.1554799104372895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57173100140, "tokens_per_second": 147704.57057288743, "data_wait_s": 1.0850002581719309e-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 sha256:113b2b55382205f6efe1d1750983df370da8519be92f2731fa39ad4920c02ed0
-size 757536
+oid sha256:9fefd3af88952ebf94d1c52445edb9a2463f358e82d002e6176fa302d0285efd
+size 759654