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": 75094.79425144794,
-      "required_pages_per_second_upper_estimate": 0.006509026439699522
+      "required_token_positions_per_second": 74117.29765017582,
+      "required_pages_per_second_upper_estimate": 0.006424299511743741
     },
     {
       "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": 9386.849281430992,
-      "required_pages_per_second_upper_estimate": 0.003744554524266392
+      "required_token_positions_per_second": 9264.662206271978,
+      "required_pages_per_second_upper_estimate": 0.003695812273125889
     },
     {
       "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": 18773.698562861984,
-      "required_pages_per_second_upper_estimate": 0.004342964756708173
+      "required_token_positions_per_second": 18529.324412543956,
+      "required_pages_per_second_upper_estimate": 0.004286433097870251
     },
     {
       "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": 28160.547844292974,
-      "required_pages_per_second_upper_estimate": 0.001714829616741391
+      "required_token_positions_per_second": 27793.986618815932,
+      "required_pages_per_second_upper_estimate": 0.0016925079613079566
     },
     {
       "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": 28160.547844292974,
-      "required_pages_per_second_upper_estimate": 0.002167876926439345
+      "required_token_positions_per_second": 27793.986618815932,
+      "required_pages_per_second_upper_estimate": 0.002139658028595706
     },
     {
       "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": 28160.547844292974,
-      "required_pages_per_second_upper_estimate": 0.34965975693523443
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27793.986618815932,
+      "required_pages_per_second_upper_estimate": 0.3451082933163134
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.111315061632979,
-    "producer_p95_seconds": 0.368932520999806,
-    "gpu_update_seconds": 0.9817990812136544,
-    "producer_batches_per_second": 8.98351027551992,
-    "gpu_batches_per_second": 1.0185383334886058,
-    "producer_headroom": 8.820002134578882,
+    "producer_batch_seconds": 0.38263307036726246,
+    "producer_p95_seconds": 0.37659361799887847,
+    "gpu_update_seconds": 0.9947475466251717,
+    "producer_batches_per_second": 2.6134698682478508,
+    "gpu_batches_per_second": 1.0052801873124975,
+    "producer_headroom": 2.5997427396183603,
     "sustainable": true,
-    "buffer_seconds": 60.871543035246575,
+    "buffer_seconds": 60.67960034413547,
     "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": 486404.0267885608,
-    "upload_bytes_per_second": 5991333.939715408,
+    "generation_bytes_per_second": 425438.3440016364,
+    "upload_bytes_per_second": 5955908.666348586,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.37098124258631,
+    "estimated_full_shards_per_hour": 91.28916492497271,
     "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": 1.2161200219998136,
-      "bytes": 5847296,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.0819334130028437,
       "bytes": 2586941,
@@ -562,6 +556,12 @@
       "bytes": 5973803,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 0.8451246749973507,
+      "bytes": 1134510,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -296,3 +296,4 @@
 {"step": 114600, "loss": 0.5923062432557344, "lr": 0.0003, "grad_norm": 0.13503638414743294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16769917267, "tokens_per_second": 150322.0356558141, "data_wait_s": 1.4290999388322234e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114700, "loss": 0.6170772984623909, "lr": 0.0003, "grad_norm": 0.14149551904141514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16784645668, "tokens_per_second": 149495.65994409827, "data_wait_s": 1.5184996300376952e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 114800, "loss": 0.5815488835796714, "lr": 0.0003, "grad_norm": 0.1341264427182585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16799325296, "tokens_per_second": 149237.73573769786, "data_wait_s": 1.0627998562995344e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 114900, "loss": 0.6156587619334459, "lr": 0.0003, "grad_norm": 0.13408592870255762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16813972431, "tokens_per_second": 149642.518388121, "data_wait_s": 1.3570999726653099e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:abbfad888d8183ae3be9e10add145cc42d83818a4391751e3579ccb41ca9bb7d
-size 635244
+oid sha256:7a045f82fb1ed260f2dd6ae17e64724464ce29b638d952b3e43baad453ab024b
+size 637362