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": 75234.6823107327,
-      "required_pages_per_second_upper_estimate": 0.006521151582135281
+      "required_token_positions_per_second": 75377.2381867214,
+      "required_pages_per_second_upper_estimate": 0.0065335079641614165
     },
     {
       "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": 9404.335288841588,
-      "required_pages_per_second_upper_estimate": 0.003751529954061588
+      "required_token_positions_per_second": 9422.154773340175,
+      "required_pages_per_second_upper_estimate": 0.0037586384128531097
     },
     {
       "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": 18808.670577683177,
-      "required_pages_per_second_upper_estimate": 0.004351054916850617
+      "required_token_positions_per_second": 18844.30954668035,
+      "required_pages_per_second_upper_estimate": 0.0043592993651036804
     },
     {
       "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": 28213.005866524763,
-      "required_pages_per_second_upper_estimate": 0.0017180240350693349
+      "required_token_positions_per_second": 28266.46432002052,
+      "required_pages_per_second_upper_estimate": 0.0017212793744123974
     },
     {
       "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": 28213.005866524763,
-      "required_pages_per_second_upper_estimate": 0.0021719152902038475
+      "required_token_positions_per_second": 28266.46432002052,
+      "required_pages_per_second_upper_estimate": 0.0021760306699364217
     },
     {
       "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": 28213.005866524763,
-      "required_pages_per_second_upper_estimate": 0.35031111000564663
+      "required_token_positions_per_second": 28266.46432002052,
+      "required_pages_per_second_upper_estimate": 0.3509748850841293
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.3386623413748566,
-    "producer_p95_seconds": 0.3197570779993839,
-    "gpu_update_seconds": 0.9799735671839507,
-    "producer_batches_per_second": 2.9527936172068383,
-    "gpu_batches_per_second": 1.0204356867232627,
-    "producer_headroom": 2.8936596942121864,
+    "producer_batch_seconds": 0.08544205853104359,
+    "producer_p95_seconds": 0.2713739539976814,
+    "gpu_update_seconds": 0.9781202094107512,
+    "producer_batches_per_second": 11.703837866180049,
+    "gpu_batches_per_second": 1.0223692245377793,
+    "producer_headroom": 11.447760344577508,
     "sustainable": true,
-    "buffer_seconds": 60.75836116540494,
+    "buffer_seconds": 60.643452983466574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441017.0792833403,
+    "generation_bytes_per_second": 442794.0212880289,
     "upload_bytes_per_second": 5662360.249152805,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.63199886202962,
+    "estimated_full_shards_per_hour": 95.01328925114298,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -193,3 +193,4 @@
 {"step": 334300, "loss": 0.6059367917478085, "lr": 0.0003, "grad_norm": 0.16072774066710382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48933096580, "tokens_per_second": 147556.20351281943, "data_wait_s": 1.0681997082429007e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 334400, "loss": 0.5995811820030212, "lr": 0.0003, "grad_norm": 0.18325775527286578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48947710891, "tokens_per_second": 147655.47456030236, "data_wait_s": 1.2196000170661137e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 334500, "loss": 0.621790231205523, "lr": 0.0003, "grad_norm": 0.15531018185033035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48962416626, "tokens_per_second": 148331.13671092314, "data_wait_s": 1.4492001355392858e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 334600, "loss": 0.5858848206698895, "lr": 0.0003, "grad_norm": 0.16661552085024728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48976990553, "tokens_per_second": 147276.2463025926, "data_wait_s": 1.2868000339949504e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5dc28d33db6293cd95861568973fa78e4856aa09ab069e3b061db817f19a8e8c
-size 415524
+oid sha256:bb38595a8dc76fb2bdb081160a222f3d20b8c0352a861542c756e2952919d7ee
+size 417642