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": 77276.51028167087,
-      "required_pages_per_second_upper_estimate": 0.005346168287493419
+      "required_token_positions_per_second": 77324.85016908222,
+      "required_pages_per_second_upper_estimate": 0.0053495125530685275
     },
     {
       "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": 9659.563785208858,
-      "required_pages_per_second_upper_estimate": 0.004044257642237379
+      "required_token_positions_per_second": 9665.606271135277,
+      "required_pages_per_second_upper_estimate": 0.004046787504913316
     },
     {
       "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": 19319.127570417717,
-      "required_pages_per_second_upper_estimate": 0.004023664793262956
+      "required_token_positions_per_second": 19331.212542270554,
+      "required_pages_per_second_upper_estimate": 0.004026181774197762
     },
     {
       "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": 28978.691355626575,
-      "required_pages_per_second_upper_estimate": 0.0017372282988701497
+      "required_token_positions_per_second": 28996.818813405833,
+      "required_pages_per_second_upper_estimate": 0.0017383150122849908
     },
     {
       "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": 28978.691355626575,
-      "required_pages_per_second_upper_estimate": 0.0022529657891074816
+      "required_token_positions_per_second": 28996.818813405833,
+      "required_pages_per_second_upper_estimate": 0.0022543751192155587
     },
     {
       "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": 28978.691355626575,
-      "required_pages_per_second_upper_estimate": 0.3717075378154023
+      "required_token_positions_per_second": 28996.818813405833,
+      "required_pages_per_second_upper_estimate": 0.3719400573800469
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05393386013284385,
-    "producer_p95_seconds": 0.19100947599872597,
-    "gpu_update_seconds": 0.954080350306495,
-    "producer_batches_per_second": 18.54122804369856,
-    "gpu_batches_per_second": 1.0481297509992251,
-    "producer_headroom": 17.68982134704453,
+    "producer_batch_seconds": 0.06619721028127401,
+    "producer_p95_seconds": 0.18908449900118285,
+    "gpu_update_seconds": 0.9534839038004319,
+    "producer_batches_per_second": 15.10637677556152,
+    "gpu_batches_per_second": 1.0487854026839494,
+    "producer_headroom": 14.40368710024258,
     "sustainable": true,
-    "buffer_seconds": 60.10706206930918,
+    "buffer_seconds": 60.06948593942721,
     "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": 487931.57431439555,
-    "upload_bytes_per_second": 4912845.983680671,
+    "generation_bytes_per_second": 488557.7219161703,
+    "upload_bytes_per_second": 4963247.776605001,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.69875738214397,
+    "estimated_full_shards_per_hour": 104.83311408151467,
     "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": 0.9362600690001273,
-      "bytes": 1025985,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.7697618709999006,
       "bytes": 7336838,
@@ -562,6 +556,12 @@
       "bytes": 5909001,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1004524960007984,
+      "bytes": 6145895,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -86,3 +86,4 @@
 {"step": 405700, "loss": 0.6061990261077881, "lr": 0.0003, "grad_norm": 0.1505809875030439, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59383779968, "tokens_per_second": 151296.8658098422, "data_wait_s": 1.0068999472423457e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 405800, "loss": 0.5753059387207031, "lr": 0.0003, "grad_norm": 0.13653449395412134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59398363644, "tokens_per_second": 151591.92799238666, "data_wait_s": 1.4731000192114152e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 405900, "loss": 0.5587671399116516, "lr": 0.0003, "grad_norm": 0.15600112342456546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59412977969, "tokens_per_second": 152177.99022793316, "data_wait_s": 1.2108999726478942e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 406000, "loss": 0.5625200271606445, "lr": 0.0003, "grad_norm": 0.16528210399480192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59427711856, "tokens_per_second": 153234.48896823038, "data_wait_s": 1.3924000086262822e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2747436211da43800a9937a7a0ddefc0d21528ad21f4f278f68bbb8604dcb454
-size 188115
+oid sha256:2311f1156addf0d81033bb85b1fdb647c2d64c53152f13752ff4c9fe262e19e2
+size 190233