Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75720.08139027432,
-      "required_pages_per_second_upper_estimate": 0.005238490925374022
+      "required_token_positions_per_second": 77296.99609113207,
+      "required_pages_per_second_upper_estimate": 0.005347585543325571
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9465.01017378429,
-      "required_pages_per_second_upper_estimate": 0.003962802107875308
+      "required_token_positions_per_second": 9662.124511391508,
+      "required_pages_per_second_upper_estimate": 0.004045329764816011
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18930.02034756858,
-      "required_pages_per_second_upper_estimate": 0.0039426240201909995
+      "required_token_positions_per_second": 19324.249022783017,
+      "required_pages_per_second_upper_estimate": 0.004024731456728866
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28395.03052135287,
-      "required_pages_per_second_upper_estimate": 0.001702238723053933
+      "required_token_positions_per_second": 28986.373534174523,
+      "required_pages_per_second_upper_estimate": 0.0017376888337440916
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28395.03052135287,
-      "required_pages_per_second_upper_estimate": 0.0022075887264953954
+      "required_token_positions_per_second": 28986.373534174523,
+      "required_pages_per_second_upper_estimate": 0.002253563044699252
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28395.03052135287,
-      "required_pages_per_second_upper_estimate": 0.3642209633195171
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28986.373534174523,
+      "required_pages_per_second_upper_estimate": 0.37180607655333464
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.8354290983906338,
-    "producer_p95_seconds": 0.2129452700000911,
-    "gpu_update_seconds": 0.9736915048993834,
-    "producer_batches_per_second": 1.1969896690531783,
-    "gpu_batches_per_second": 1.0270193330929134,
-    "producer_headroom": 1.165498672209404,
+    "producer_batch_seconds": 0.07694203342182249,
+    "producer_p95_seconds": 0.1897517980005432,
+    "gpu_update_seconds": 0.9538274930254176,
+    "producer_batches_per_second": 12.996797141006901,
+    "gpu_batches_per_second": 1.0484076075728632,
+    "producer_headroom": 12.396702434366528,
     "sustainable": true,
-    "buffer_seconds": 60.36887330376177,
+    "buffer_seconds": 60.09113206060131,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 259970.99361244505,
-    "upload_bytes_per_second": 5162817.70327222,
+    "generation_bytes_per_second": 259366.02390744665,
+    "upload_bytes_per_second": 5319750.378231893,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.78372341423048,
+    "estimated_full_shards_per_hour": 55.65391099851179,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 947666,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.5761228700002903,
+      "bytes": 18281686,
+      "pages": 39,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -74,3 +74,4 @@
 {"step": 892400, "loss": 0.5268421769142151, "lr": 3e-05, "grad_norm": 0.3321612467374445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130621985156, "tokens_per_second": 151539.6313387353, "data_wait_s": 1.5497000276809558e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 892500, "loss": 0.5253483057022095, "lr": 3e-05, "grad_norm": 0.3097779684051168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130636639697, "tokens_per_second": 150566.48840920988, "data_wait_s": 9.980999493564013e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 892600, "loss": 0.48416364192962646, "lr": 3e-05, "grad_norm": 0.3105130838611581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130651373828, "tokens_per_second": 153575.1050105071, "data_wait_s": 1.1643999641819391e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 892700, "loss": 0.5476626753807068, "lr": 3e-05, "grad_norm": 0.3149015245871685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130666031796, "tokens_per_second": 153026.12313189296, "data_wait_s": 1.1962999451498035e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:966793fcbb6b4037b0c5ee4cd4ccd341b82dfdc3f48732579c92c9322e53245f
-size 161916
+oid sha256:a8270754e95f7a7146e9876838a199148d7ae2fbe288fbf547217865d7778d0f
+size 164034