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": 76247.66767374524,
-      "required_pages_per_second_upper_estimate": 0.006608954586013276
+      "required_token_positions_per_second": 76047.57103533958,
+      "required_pages_per_second_upper_estimate": 0.006591610716536568
     },
     {
       "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": 9530.958459218155,
-      "required_pages_per_second_upper_estimate": 0.0038020418299099068
+      "required_token_positions_per_second": 9505.946379417448,
+      "required_pages_per_second_upper_estimate": 0.0037920641373134865
     },
     {
       "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": 19061.91691843631,
-      "required_pages_per_second_upper_estimate": 0.004409639000800482
+      "required_token_positions_per_second": 19011.892758834896,
+      "required_pages_per_second_upper_estimate": 0.0043980667918718346
     },
     {
       "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": 28592.87537765446,
-      "required_pages_per_second_upper_estimate": 0.0017411560952758376
+      "required_token_positions_per_second": 28517.839138252344,
+      "required_pages_per_second_upper_estimate": 0.0017365867819809711
     },
     {
       "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": 28592.87537765446,
-      "required_pages_per_second_upper_estimate": 0.0022011586967167203
+      "required_token_positions_per_second": 28517.839138252344,
+      "required_pages_per_second_upper_estimate": 0.0021953821992939404
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28592.87537765446,
-      "required_pages_per_second_upper_estimate": 0.35502781799240674
+      "required_token_positions_per_second": 28517.839138252344,
+      "required_pages_per_second_upper_estimate": 0.354096119029171
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14486512628178616,
-    "producer_p95_seconds": 0.35939582900027744,
-    "gpu_update_seconds": 0.9669541672471007,
-    "producer_batches_per_second": 6.902972617818576,
-    "gpu_batches_per_second": 1.034175180036692,
-    "producer_headroom": 6.6748581391923,
+    "producer_batch_seconds": 0.4566149622730791,
+    "producer_p95_seconds": 0.36624851200031117,
+    "gpu_update_seconds": 0.9694984204786544,
+    "producer_batches_per_second": 2.1900289798255645,
+    "gpu_batches_per_second": 1.0314611956833168,
+    "producer_headroom": 2.1232296367433636,
     "sustainable": true,
-    "buffer_seconds": 60.91811253656734,
+    "buffer_seconds": 60.108902069676574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373906.82441998215,
-    "upload_bytes_per_second": 4833210.72004389,
+    "generation_bytes_per_second": 345885.198030879,
+    "upload_bytes_per_second": 4740309.041704067,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.2317004151306,
+    "estimated_full_shards_per_hour": 74.21891170210627,
     "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": 1.1992983269992692,
-      "bytes": 13571269,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.1989007740012312,
       "bytes": 5658429,
@@ -562,6 +556,12 @@
       "bytes": 877422,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.6349642249988392,
+      "bytes": 7479267,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -390,3 +390,4 @@
 {"step": 194000, "loss": 0.7089795097708702, "lr": 0.0003, "grad_norm": 0.18382650708824108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28393808966, "tokens_per_second": 149189.65971278542, "data_wait_s": 9.483002941124141e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 194100, "loss": 0.5605682022869587, "lr": 0.0003, "grad_norm": 0.1346691769896389, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28408327117, "tokens_per_second": 149352.59890534455, "data_wait_s": 8.06100433692336e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 194200, "loss": 0.6196492426097393, "lr": 0.0003, "grad_norm": 0.1536003217165317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28423003726, "tokens_per_second": 141485.18286606195, "data_wait_s": 1.170699397334829e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 194300, "loss": 0.6588104479014874, "lr": 0.0003, "grad_norm": 0.15547511138598508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28437767900, "tokens_per_second": 151689.48828735013, "data_wait_s": 1.3245997251942754e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bcddf2a0d13ffa17f49434ac64a3bb2dd43249c1ad6d632605092e0729053642
-size 835902
+oid sha256:8817e95e63a4cf1070797fea6045f467bdb6eccdad21d24cb7b8682c03fb68ce
+size 838020