Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76619.31638446944,
-      "required_pages_per_second_upper_estimate": 0.005300702088785092
+      "required_token_positions_per_second": 77504.2232919795,
+      "required_pages_per_second_upper_estimate": 0.005361922001913549
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9577.41454805868,
-      "required_pages_per_second_upper_estimate": 0.004009863472113743
+      "required_token_positions_per_second": 9688.027911497438,
+      "required_pages_per_second_upper_estimate": 0.004056174977515859
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19154.82909611736,
-      "required_pages_per_second_upper_estimate": 0.003989445753908329
+      "required_token_positions_per_second": 19376.055822994877,
+      "required_pages_per_second_upper_estimate": 0.0040355214469757995
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28732.24364417604,
-      "required_pages_per_second_upper_estimate": 0.0017224541348725567
+      "required_token_positions_per_second": 29064.083734492313,
+      "required_pages_per_second_upper_estimate": 0.0017423474415965378
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28732.24364417604,
-      "required_pages_per_second_upper_estimate": 0.0022338055635511235
+      "required_token_positions_per_second": 29064.083734492313,
+      "required_pages_per_second_upper_estimate": 0.002259604670962913
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28732.24364417604,
-      "required_pages_per_second_upper_estimate": 0.36854637118785083
+      "required_token_positions_per_second": 29064.083734492313,
+      "required_pages_per_second_upper_estimate": 0.372802859564299
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07736619592174065,
+    "producer_batch_seconds": 0.0943874502889912,
     "producer_p95_seconds": 0.2163933880001423,
-    "gpu_update_seconds": 0.9622638712937472,
-    "producer_batches_per_second": 12.925541809132563,
-    "gpu_batches_per_second": 1.039215988287617,
-    "producer_headroom": 12.437781899825085,
+    "gpu_update_seconds": 0.9512771932730241,
+    "producer_batches_per_second": 10.594628808578317,
+    "gpu_batches_per_second": 1.0512183063690796,
+    "producer_headroom": 10.078428756793905,
     "sustainable": true,
-    "buffer_seconds": 60.62262389150607,
+    "buffer_seconds": 60.88174036947354,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320114.27023038676,
+    "generation_bytes_per_second": 357789.73475888773,
     "upload_bytes_per_second": 5321536.556669298,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.68907051261618,
+    "estimated_full_shards_per_hour": 76.77334816050505,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -175,3 +175,4 @@
 {"step": 574600, "loss": 0.5650253295898438, "lr": 0.0003, "grad_norm": 0.1595260223471998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84104123104, "tokens_per_second": 153091.42426447838, "data_wait_s": 2.3594999220222235e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 574700, "loss": 0.6336092352867126, "lr": 0.0003, "grad_norm": 0.15673328562349978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84118631127, "tokens_per_second": 151586.2848648544, "data_wait_s": 9.937000868376344e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 574800, "loss": 0.7473382353782654, "lr": 0.0003, "grad_norm": 0.3193019051904646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84133290730, "tokens_per_second": 152766.53100402327, "data_wait_s": 1.2799999240087345e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 574900, "loss": 0.5708333849906921, "lr": 0.0003, "grad_norm": 0.1551997033269296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84148032987, "tokens_per_second": 154211.103268809, "data_wait_s": 8.013001206563786e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b4a01d5ea7dce57a78551f4bd68a15ecd1403bf428a40a664e7e1fcdcf6dfc80
-size 377400
+oid sha256:1d5ff27c23b7a6efa2aab93433dcd3932b979f3a7d7fea3e8d22a5761393ea2b
+size 379518