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": 74519.3417157271,
-      "required_pages_per_second_upper_estimate": 0.006459147672374358
+      "required_token_positions_per_second": 75551.06748499216,
+      "required_pages_per_second_upper_estimate": 0.006548575047169197
     },
     {
       "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": 9314.917714465888,
-      "required_pages_per_second_upper_estimate": 0.003715859946731246
+      "required_token_positions_per_second": 9443.88343562402,
+      "required_pages_per_second_upper_estimate": 0.003767306301110587
     },
     {
       "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": 18629.835428931776,
-      "required_pages_per_second_upper_estimate": 0.004309684552578102
+      "required_token_positions_per_second": 18887.76687124804,
+      "required_pages_per_second_upper_estimate": 0.004369352452319625
     },
     {
       "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": 27944.753143397666,
-      "required_pages_per_second_upper_estimate": 0.0017016888516441623
+      "required_token_positions_per_second": 28331.650306872063,
+      "required_pages_per_second_upper_estimate": 0.0017252488590072147
     },
     {
       "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": 27944.753143397666,
-      "required_pages_per_second_upper_estimate": 0.002151264453006459
+      "required_token_positions_per_second": 28331.650306872063,
+      "required_pages_per_second_upper_estimate": 0.002181048867650614
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27944.753143397666,
-      "required_pages_per_second_upper_estimate": 0.3469803089685196
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28331.650306872063,
+      "required_pages_per_second_upper_estimate": 0.3517842768773615
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08579358607812537,
-    "producer_p95_seconds": 0.2775592729994969,
-    "gpu_update_seconds": 0.9893807205282907,
-    "producer_batches_per_second": 11.655882982783583,
-    "gpu_batches_per_second": 1.0107332589481215,
-    "producer_headroom": 11.532105903899863,
+    "producer_batch_seconds": 0.0969070824689311,
+    "producer_p95_seconds": 0.30215932999999495,
+    "gpu_update_seconds": 0.975869732279371,
+    "producer_batches_per_second": 10.319163207917287,
+    "gpu_batches_per_second": 1.0247269352890649,
+    "producer_headroom": 10.070159037057378,
     "sustainable": true,
-    "buffer_seconds": 60.352223952225735,
+    "buffer_seconds": 60.503923401321,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362506.82181637763,
+    "generation_bytes_per_second": 363164.37556073116,
     "upload_bytes_per_second": 6019438.499153285,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.78552523487565,
+    "estimated_full_shards_per_hour": 77.92662096134616,
     "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": [

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

@@ -160,3 +160,4 @@
 {"step": 101000, "loss": 0.6123548597097397, "lr": 0.0003, "grad_norm": 0.131153111294977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14779595813, "tokens_per_second": 149193.15899403312, "data_wait_s": 1.1594998795771971e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 101100, "loss": 0.6251995395869017, "lr": 0.0003, "grad_norm": 0.13753801862563045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14794354040, "tokens_per_second": 149332.41022912136, "data_wait_s": 8.861999958753586e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 101200, "loss": 0.6119210012257099, "lr": 0.0003, "grad_norm": 0.12931871012234633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14809187848, "tokens_per_second": 151153.94490864305, "data_wait_s": 1.4155997632769868e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 101300, "loss": 0.5754358042031527, "lr": 0.0003, "grad_norm": 0.17327013163768856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14823943277, "tokens_per_second": 148975.148350734, "data_wait_s": 2.0988998585380614e-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:e254ac83b073d7b9a65d2c7d9b1e1e1e8f35aca7bf3cb3f38e05a5c60fbc03ef
-size 345630
+oid sha256:2abcdf6830fc379792af5daec282da2d84d1fa8397f2cef5852cce8ae8427432
+size 347748