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": 76844.57809205123,
-      "required_pages_per_second_upper_estimate": 0.0053162862163947736
+      "required_token_positions_per_second": 77335.07447722883,
+      "required_pages_per_second_upper_estimate": 0.005350219894429774
     },
     {
       "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": 9605.572261506404,
-      "required_pages_per_second_upper_estimate": 0.004021652518734385
+      "required_token_positions_per_second": 9666.884309653604,
+      "required_pages_per_second_upper_estimate": 0.00404732259295246
     },
     {
       "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": 19211.14452301281,
-      "required_pages_per_second_upper_estimate": 0.004001174772192973
+      "required_token_positions_per_second": 19333.768619307208,
+      "required_pages_per_second_upper_estimate": 0.004026714137636212
     },
     {
       "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": 28816.716784519213,
-      "required_pages_per_second_upper_estimate": 0.0017275181706531129
+      "required_token_positions_per_second": 29000.65292896081,
+      "required_pages_per_second_upper_estimate": 0.0017385448616581564
     },
     {
       "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": 28816.716784519213,
-      "required_pages_per_second_upper_estimate": 0.0022403729786547285
+      "required_token_positions_per_second": 29000.65292896081,
+      "required_pages_per_second_upper_estimate": 0.0022546732048354663
     },
     {
       "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": 28816.716784519213,
-      "required_pages_per_second_upper_estimate": 0.36962990193198153
+      "required_token_positions_per_second": 29000.65292896081,
+      "required_pages_per_second_upper_estimate": 0.37198923729763356
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11217400132034072,
-    "producer_p95_seconds": 0.3455148119974183,
-    "gpu_update_seconds": 0.9594430971002544,
-    "producer_batches_per_second": 8.914721666603045,
-    "gpu_batches_per_second": 1.042271295736372,
-    "producer_headroom": 8.553168165592368,
+    "producer_batch_seconds": 0.19726435821058885,
+    "producer_p95_seconds": 0.3825599379997584,
+    "gpu_update_seconds": 0.9533578456914666,
+    "producer_batches_per_second": 5.069339484695221,
+    "gpu_batches_per_second": 1.0489240787384553,
+    "producer_headroom": 4.832894570207726,
     "sustainable": true,
-    "buffer_seconds": 60.44491511731603,
+    "buffer_seconds": 60.0615442785624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318894.3651239485,
-    "upload_bytes_per_second": 5013614.90481619,
+    "generation_bytes_per_second": 338201.7555751353,
+    "upload_bytes_per_second": 5059497.530243959,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.427307274712,
+    "estimated_full_shards_per_hour": 72.57022381248993,
     "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": 1.1377382460013905,
-      "bytes": 6053085,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 0.9952570979985467,
       "bytes": 952538,
@@ -562,6 +556,12 @@
       "bytes": 923281,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.7383342659959453,
+      "bytes": 12982783,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -393,3 +393,4 @@
 {"step": 964300, "loss": 0.5127605199813843, "lr": 3e-05, "grad_norm": 0.34766408444710684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141147315034, "tokens_per_second": 152906.0295405701, "data_wait_s": 1.1350006388965994e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 964400, "loss": 0.5766127109527588, "lr": 3e-05, "grad_norm": 0.35633526615795746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141162020865, "tokens_per_second": 152861.3945916866, "data_wait_s": 1.4850003935862333e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 964500, "loss": 0.5489728450775146, "lr": 3e-05, "grad_norm": 0.33473212707880906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141176557068, "tokens_per_second": 151519.9288141041, "data_wait_s": 1.1498996173031628e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 964600, "loss": 0.530150294303894, "lr": 3e-05, "grad_norm": 0.34140706599839477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141190980952, "tokens_per_second": 149598.39953637886, "data_wait_s": 1.2651995348278433e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9ca1099e53d1fd6a9ad43a6c8e3f73ed14be8543ca87cd6c9d92c3aa7e67c11
-size 842256
+oid sha256:55cad219860adfb979fe55530d9011a8056272f477b95c8e04a284aadd0c466b
+size 844374