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": 75060.141346076,
-      "required_pages_per_second_upper_estimate": 0.006506022813688867
+      "required_token_positions_per_second": 74239.71051296088,
+      "required_pages_per_second_upper_estimate": 0.006434909948437386
     },
     {
       "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": 9382.5176682595,
-      "required_pages_per_second_upper_estimate": 0.0037428265790088954
+      "required_token_positions_per_second": 9279.96381412011,
+      "required_pages_per_second_upper_estimate": 0.0037019163132759334
     },
     {
       "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": 18765.035336519,
-      "required_pages_per_second_upper_estimate": 0.0043409606717613
+      "required_token_positions_per_second": 18559.92762824022,
+      "required_pages_per_second_upper_estimate": 0.004293512613223381
     },
     {
       "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": 28147.553004778496,
-      "required_pages_per_second_upper_estimate": 0.0017140382991936132
+      "required_token_positions_per_second": 27839.891442360327,
+      "required_pages_per_second_upper_estimate": 0.0016953033242178147
     },
     {
       "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": 28147.553004778496,
-      "required_pages_per_second_upper_estimate": 0.002166876547721502
+      "required_token_positions_per_second": 27839.891442360327,
+      "required_pages_per_second_upper_estimate": 0.002143191908984838
     },
     {
       "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": 28147.553004778496,
-      "required_pages_per_second_upper_estimate": 0.3494984045193948
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27839.891442360327,
+      "required_pages_per_second_upper_estimate": 0.34567827759117337
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.055946034054663585,
-    "producer_p95_seconds": 0.20531754300009197,
-    "gpu_update_seconds": 0.9822523469555705,
-    "producer_batches_per_second": 17.874367985100122,
-    "gpu_batches_per_second": 1.018068323378852,
-    "producer_headroom": 17.55713990371211,
+    "producer_batch_seconds": 0.11302162930473258,
+    "producer_p95_seconds": 0.2140635759997167,
+    "gpu_update_seconds": 0.9931073207394641,
+    "producer_batches_per_second": 8.8478639544628,
+    "gpu_batches_per_second": 1.0069405180251856,
+    "producer_headroom": 8.78687846608383,
     "sustainable": true,
-    "buffer_seconds": 60.899645511245375,
+    "buffer_seconds": 60.57954656510731,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332789.31418504246,
-    "upload_bytes_per_second": 5932626.484913232,
+    "generation_bytes_per_second": 362290.61797311145,
+    "upload_bytes_per_second": 5902263.000178364,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.40883988536315,
+    "estimated_full_shards_per_hour": 77.73913292307861,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 1212988,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0933355659999506,
+      "bytes": 5690232,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -74,3 +74,4 @@
 {"step": 122400, "loss": 0.6381675526499748, "lr": 0.0003, "grad_norm": 0.17643904933300172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17912108499, "tokens_per_second": 150145.00363960842, "data_wait_s": 1.4076999832468573e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 122500, "loss": 0.5423185117542744, "lr": 0.0003, "grad_norm": 0.18062500836188222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17926705528, "tokens_per_second": 149139.726335044, "data_wait_s": 1.0955999641737435e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 122600, "loss": 0.6145331133157015, "lr": 0.0003, "grad_norm": 0.1591392239095421, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17941202218, "tokens_per_second": 147406.74475259118, "data_wait_s": 1.597299979039235e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 122700, "loss": 0.6094257552176714, "lr": 0.0003, "grad_norm": 0.13699610787656766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17955682394, "tokens_per_second": 148236.24361761165, "data_wait_s": 1.1531999916769564e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e1b0721b8f07979c6aaae432c69a90109a9931d7ad3cf048fd550160b8d471c
-size 161916
+oid sha256:33c8f5b30578159d6a83ef714012e2e854c7c8249a957a0a62e1452502a11d2f
+size 164034