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": 75368.18777639217,
-      "required_pages_per_second_upper_estimate": 0.006532723497532133
+      "required_token_positions_per_second": 73314.96087456509,
+      "required_pages_per_second_upper_estimate": 0.006354754993537769
     },
     {
       "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": 9421.02347204902,
-      "required_pages_per_second_upper_estimate": 0.0037581871198536063
+      "required_token_positions_per_second": 9164.370109320636,
+      "required_pages_per_second_upper_estimate": 0.003655804256151522
     },
     {
       "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": 18842.04694409804,
-      "required_pages_per_second_upper_estimate": 0.004358775951816709
+      "required_token_positions_per_second": 18328.740218641273,
+      "required_pages_per_second_upper_estimate": 0.004240031474772108
     },
     {
       "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": 28263.070416147064,
-      "required_pages_per_second_upper_estimate": 0.0017210727034729413
+      "required_token_positions_per_second": 27493.11032796191,
+      "required_pages_per_second_upper_estimate": 0.001674186173770846
     },
     {
       "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": 28263.070416147064,
-      "required_pages_per_second_upper_estimate": 0.0021757693978213155
+      "required_token_positions_per_second": 27493.11032796191,
+      "required_pages_per_second_upper_estimate": 0.0021164957388468833
     },
     {
       "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": 28263.070416147064,
-      "required_pages_per_second_upper_estimate": 0.3509327441566865
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27493.11032796191,
+      "required_pages_per_second_upper_estimate": 0.3413724167520756
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1606457544066302,
-    "producer_p95_seconds": 0.3911977770039812,
-    "gpu_update_seconds": 0.9782376646595458,
-    "producer_batches_per_second": 6.2248766155922,
-    "gpu_batches_per_second": 1.0222464704914302,
-    "producer_headroom": 6.089408763230731,
+    "producer_batch_seconds": 0.4578662695237199,
+    "producer_p95_seconds": 0.39385348799987696,
+    "gpu_update_seconds": 1.0056337631570393,
+    "producer_batches_per_second": 2.18404382799418,
+    "gpu_batches_per_second": 0.9943977983203816,
+    "producer_headroom": 2.1963482136456927,
     "sustainable": true,
-    "buffer_seconds": 60.65073520889184,
+    "buffer_seconds": 60.33802578942236,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362750.0021895322,
-    "upload_bytes_per_second": 5470118.08714139,
+    "generation_bytes_per_second": 309390.56707928464,
+    "upload_bytes_per_second": 5499902.460362616,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.83770608200525,
+    "estimated_full_shards_per_hour": 66.38801345142274,
     "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": 1.7747774790004769,
-      "bytes": 7374031,
-      "pages": 33,
-      "shards": 3
-    },
-    {
-      "seconds": 0.9477112800013856,
-      "bytes": 5792070,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.0662493220006581,
       "bytes": 5942385,
@@ -562,6 +550,18 @@
       "bytes": 902486,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1300200109981233,
+      "bytes": 7391590,
+      "pages": 31,
+      "shards": 3
+    },
+    {
+      "seconds": 0.8869208329997491,
+      "bytes": 4199574,
+      "pages": 23,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -348,3 +348,4 @@
 {"step": 149800, "loss": 0.662655808031559, "lr": 0.0003, "grad_norm": 0.16492261881099443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21922915407, "tokens_per_second": 151305.0044205704, "data_wait_s": 1.0435003787279129e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 149900, "loss": 0.6620606575161219, "lr": 0.0003, "grad_norm": 0.15283104603146141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21937642348, "tokens_per_second": 149515.8032362199, "data_wait_s": 1.1324002116452903e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 150000, "loss": 0.6477003693580627, "lr": 0.0003, "grad_norm": 0.14979309409476083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21952273112, "tokens_per_second": 148775.43239117687, "data_wait_s": 1.6620004316791892e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 150100, "loss": 0.6310819704085588, "lr": 0.0003, "grad_norm": 0.14523803354176887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21966845945, "tokens_per_second": 135041.90131722114, "data_wait_s": 9.889001375995576e-06, "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 sha256:80ec4a3cd608f469d84e636907166f69ede2c5640422d677ea68ddf05f5048cc
-size 746946
+oid sha256:e219f5d953dd18aeb2f54d72b87873798b3cdcfd6d1eb58d16f711e493e2e508
+size 749064