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": 75187.91206054503,
-      "required_pages_per_second_upper_estimate": 0.006517097655387129
+      "required_token_positions_per_second": 74832.05755850897,
+      "required_pages_per_second_upper_estimate": 0.006486253089055633
     },
     {
       "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": 9398.489007568129,
-      "required_pages_per_second_upper_estimate": 0.0037491977850519103
+      "required_token_positions_per_second": 9354.007194813621,
+      "required_pages_per_second_upper_estimate": 0.00373145332488177
     },
     {
       "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": 18796.978015136257,
-      "required_pages_per_second_upper_estimate": 0.00434835004828746
+      "required_token_positions_per_second": 18708.014389627242,
+      "required_pages_per_second_upper_estimate": 0.004327769879232282
     },
     {
       "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": 28195.467022704386,
-      "required_pages_per_second_upper_estimate": 0.0017169560114998759
+      "required_token_positions_per_second": 28062.021584440867,
+      "required_pages_per_second_upper_estimate": 0.001708829884443733
     },
     {
       "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": 28195.467022704386,
-      "required_pages_per_second_upper_estimate": 0.002170565101455928
+      "required_token_positions_per_second": 28062.021584440867,
+      "required_pages_per_second_upper_estimate": 0.002160292102217786
     },
     {
       "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": 28195.467022704386,
-      "required_pages_per_second_upper_estimate": 0.35009333626413186
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28062.021584440867,
+      "required_pages_per_second_upper_estimate": 0.348436390534051
     }
   ],
   "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": 0.07750084701567062,
-    "producer_p95_seconds": 0.31806549199973233,
-    "gpu_update_seconds": 0.9805831546516488,
-    "producer_batches_per_second": 12.903084785612739,
-    "gpu_batches_per_second": 1.0198013246059168,
-    "producer_headroom": 12.652547583813833,
+    "producer_batch_seconds": 0.3227005402888494,
+    "producer_p95_seconds": 0.32684306999726687,
+    "gpu_update_seconds": 0.9852461953535656,
+    "producer_batches_per_second": 3.0988482359059564,
+    "gpu_batches_per_second": 1.0149747390205752,
+    "producer_headroom": 3.053128434404452,
     "sustainable": true,
-    "buffer_seconds": 60.79615558840223,
+    "buffer_seconds": 60.1000179165675,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348859.0339204763,
-    "upload_bytes_per_second": 5066114.178175092,
+    "generation_bytes_per_second": 328958.5480229098,
+    "upload_bytes_per_second": 4987956.395588445,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.85702765665738,
+    "estimated_full_shards_per_hour": 70.58684664264173,
     "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": 1.42962039700069,
-      "bytes": 6308465,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0526814939985343,
       "bytes": 2354967,
@@ -562,6 +556,12 @@
       "bytes": 7307496,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.6164238569981535,
+      "bytes": 892499,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -243,3 +243,4 @@
 {"step": 261400, "loss": 0.609815489500761, "lr": 0.0003, "grad_norm": 0.16562488191582922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38261832378, "tokens_per_second": 147960.23528569652, "data_wait_s": 1.4483000995824113e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 261500, "loss": 0.5521674687042832, "lr": 0.0003, "grad_norm": 0.14959994957872588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38276317946, "tokens_per_second": 147578.46847866013, "data_wait_s": 1.921099828905426e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 261600, "loss": 0.6089760884642601, "lr": 0.0003, "grad_norm": 0.12999718473535907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38291041851, "tokens_per_second": 149931.84472369438, "data_wait_s": 9.758998203324154e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 261700, "loss": 0.5810473840683699, "lr": 0.0003, "grad_norm": 0.14975278904965478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38305751328, "tokens_per_second": 149156.38515814557, "data_wait_s": 1.4618999557569623e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9128405168e63e282b6130b9342afbab5a862355c12b77c4ada1c000ce86c4ca
-size 522990
+oid sha256:785bce5a28eb0b433602ed83e32cc29d2987316118e5287217a6bc983f768a03
+size 525108