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": 73184.09904487277,
-      "required_pages_per_second_upper_estimate": 0.006343412221806314
+      "required_token_positions_per_second": 74719.95379552428,
+      "required_pages_per_second_upper_estimate": 0.00647653621900451
     },
     {
       "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": 9148.012380609096,
-      "required_pages_per_second_upper_estimate": 0.00364927891359865
+      "required_token_positions_per_second": 9339.994224440536,
+      "required_pages_per_second_upper_estimate": 0.0037258633414873686
     },
     {
       "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": 18296.024761218192,
-      "required_pages_per_second_upper_estimate": 0.004232463329469664
+      "required_token_positions_per_second": 18679.98844888107,
+      "required_pages_per_second_upper_estimate": 0.004321286571080367
     },
     {
       "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": 27444.037141827288,
-      "required_pages_per_second_upper_estimate": 0.0016711978742023571
+      "required_token_positions_per_second": 28019.982673321607,
+      "required_pages_per_second_upper_estimate": 0.0017062699353176908
     },
     {
       "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": 27444.037141827288,
-      "required_pages_per_second_upper_estimate": 0.002112717949134967
+      "required_token_positions_per_second": 28019.982673321607,
+      "required_pages_per_second_upper_estimate": 0.0021570558304686714
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27444.037141827288,
-      "required_pages_per_second_upper_estimate": 0.34076309201767246
+      "required_token_positions_per_second": 28019.982673321607,
+      "required_pages_per_second_upper_estimate": 0.34791440795313466
     }
   ],
   "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": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09436940491403334,
-    "producer_p95_seconds": 0.3166313209985674,
-    "gpu_update_seconds": 1.0074319553321787,
-    "producer_batches_per_second": 10.596654719937664,
-    "gpu_batches_per_second": 0.9926228711598412,
-    "producer_headroom": 10.675408584486762,
+    "producer_batch_seconds": 0.10350516982819613,
+    "producer_p95_seconds": 0.31090817299991613,
+    "gpu_update_seconds": 0.9867243788956451,
+    "producer_batches_per_second": 9.661353163903387,
+    "gpu_batches_per_second": 1.0134542344228012,
+    "producer_headroom": 9.533092699944044,
     "sustainable": true,
-    "buffer_seconds": 60.44591731993072,
+    "buffer_seconds": 60.190187112634355,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454794.5943717576,
-    "upload_bytes_per_second": 4024670.2377236146,
+    "generation_bytes_per_second": 478269.92669255525,
+    "upload_bytes_per_second": 4033005.262052912,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.58833287586732,
+    "estimated_full_shards_per_hour": 102.62559271414274,
     "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": 0.9415913409993664,
-      "bytes": 652793,
-      "pages": 19,
-      "shards": 1
-    },
     {
       "seconds": 1.3964870420004445,
       "bytes": 13572779,
@@ -562,6 +556,12 @@
       "bytes": 5888359,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.8503742749999219,
+      "bytes": 1173601,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -217,3 +217,4 @@
 {"step": 136700, "loss": 0.6210627965629101, "lr": 0.0003, "grad_norm": 0.15209265436022484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20005535906, "tokens_per_second": 147356.34552390876, "data_wait_s": 1.413200152455829e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 136800, "loss": 0.5940307341516018, "lr": 0.0003, "grad_norm": 0.15701820478888792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20020180662, "tokens_per_second": 148697.8069655777, "data_wait_s": 6.69799919705838e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 136900, "loss": 0.6081352643668652, "lr": 0.0003, "grad_norm": 0.12146628698002741, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20034814272, "tokens_per_second": 149217.33587462825, "data_wait_s": 8.095998055068776e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 137000, "loss": 0.5834247525781393, "lr": 0.0003, "grad_norm": 0.15340312132563966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20049573944, "tokens_per_second": 150249.84502143835, "data_wait_s": 1.5495999832637608e-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 sha256:0b3209beb3c189a166d2ab86dc49e86c18afaeab3cc626c4175183bb9c272e7c
-size 467139
+oid sha256:3849aae1d302883a84064817aa3651db96d24a1bb9120a62b00ab3e6e86d35e9
+size 469257