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": 74931.84959177274,
-      "required_pages_per_second_upper_estimate": 0.0064949028095783404
+      "required_token_positions_per_second": 75203.3641552847,
+      "required_pages_per_second_upper_estimate": 0.006518437003796187
     },
     {
       "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": 9366.481198971593,
-      "required_pages_per_second_upper_estimate": 0.003736429391643367
+      "required_token_positions_per_second": 9400.420519410587,
+      "required_pages_per_second_upper_estimate": 0.0037499682940045423
     },
     {
       "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": 18732.962397943185,
-      "required_pages_per_second_upper_estimate": 0.0043335411618862256
+      "required_token_positions_per_second": 18800.841038821174,
+      "required_pages_per_second_upper_estimate": 0.00434924369083007
     },
     {
       "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": 28099.443596914778,
-      "required_pages_per_second_upper_estimate": 0.0017111086886652685
+      "required_token_positions_per_second": 28201.261558231763,
+      "required_pages_per_second_upper_estimate": 0.0017173088683119148
     },
     {
       "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": 28099.443596914778,
-      "required_pages_per_second_upper_estimate": 0.002163172952328789
+      "required_token_positions_per_second": 28201.261558231763,
+      "required_pages_per_second_upper_estimate": 0.0021710111808411264
     },
     {
       "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": 28099.443596914778,
-      "required_pages_per_second_upper_estimate": 0.34890104668555794
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28201.261558231763,
+      "required_pages_per_second_upper_estimate": 0.35016528500231897
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.23628611993753168,
-    "producer_p95_seconds": 0.2043264320000162,
-    "gpu_update_seconds": 0.9839340734503247,
-    "producer_batches_per_second": 4.2321571841138015,
-    "gpu_batches_per_second": 1.016328255096744,
-    "producer_headroom": 4.164163657647149,
+    "producer_batch_seconds": 0.07806653850782652,
+    "producer_p95_seconds": 0.17480907999924966,
+    "gpu_update_seconds": 0.9803816734549498,
+    "producer_batches_per_second": 12.809585503778235,
+    "gpu_batches_per_second": 1.0200109070541001,
+    "producer_headroom": 12.558282872458372,
     "sustainable": true,
-    "buffer_seconds": 60.0199784804698,
+    "buffer_seconds": 60.783663754206884,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443998.2543192281,
-    "upload_bytes_per_second": 5617608.209122828,
+    "generation_bytes_per_second": 444260.6866708628,
+    "upload_bytes_per_second": 5666234.76505605,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.2716896265281,
+    "estimated_full_shards_per_hour": 95.32800150007641,
     "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": [
@@ -418,6 +418,12 @@
       "bytes": 1154123,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.3542424540000866,
+      "bytes": 9851710,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -73,3 +73,4 @@
 {"step": 362300, "loss": 0.581356992945075, "lr": 0.0003, "grad_norm": 0.1601125700377772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53030612571, "tokens_per_second": 149150.71362877195, "data_wait_s": 7.561000529676676e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 362400, "loss": 0.6729161273688078, "lr": 0.0003, "grad_norm": 0.18195817486300464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53045348843, "tokens_per_second": 150079.46739753152, "data_wait_s": 8.346999493369367e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 362500, "loss": 0.558849785476923, "lr": 0.0003, "grad_norm": 0.13784731743430725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53060022456, "tokens_per_second": 149752.11108908063, "data_wait_s": 1.276300008612452e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 362600, "loss": 0.5897896364331245, "lr": 0.0003, "grad_norm": 0.15633946977455013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53074429274, "tokens_per_second": 146945.11608346397, "data_wait_s": 1.0820000170497224e-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:d922ea7f74a6c6dac3aa3404965153a6ebf11efcdee46b18da2394676b9aca40
-size 159798
+oid sha256:05c1d438c44b93aacea6f72f573855bb57b364826355547da75b62b4737d2fcb
+size 161916