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": 77342.64494941375,
-      "required_pages_per_second_upper_estimate": 0.005350743637261442
+      "required_token_positions_per_second": 77267.78512016054,
+      "required_pages_per_second_upper_estimate": 0.00534556466057497
     },
     {
       "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": 9667.830618676719,
-      "required_pages_per_second_upper_estimate": 0.0040477187927792585
+      "required_token_positions_per_second": 9658.473140020067,
+      "required_pages_per_second_upper_estimate": 0.004043801011872093
     },
     {
       "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": 19335.661237353437,
-      "required_pages_per_second_upper_estimate": 0.0040271083200635474
+      "required_token_positions_per_second": 19316.946280040134,
+      "required_pages_per_second_upper_estimate": 0.004023210488001801
     },
     {
       "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": 29003.491856030156,
-      "required_pages_per_second_upper_estimate": 0.0017387150510011727
+      "required_token_positions_per_second": 28975.419420060203,
+      "required_pages_per_second_upper_estimate": 0.001737032151328902
     },
     {
       "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": 29003.491856030156,
-      "required_pages_per_second_upper_estimate": 0.0022548939189279875
+      "required_token_positions_per_second": 28975.419420060203,
+      "required_pages_per_second_upper_estimate": 0.00225271141050892
     },
     {
       "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": 29003.491856030156,
-      "required_pages_per_second_upper_estimate": 0.37202565200587673
+      "required_token_positions_per_second": 28975.419420060203,
+      "required_pages_per_second_upper_estimate": 0.37166556893908753
     }
   ],
   "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": 0.26844955488240885,
-    "producer_p95_seconds": 0.3336269299979904,
-    "gpu_update_seconds": 0.9532645288795344,
-    "producer_batches_per_second": 3.7250946474395836,
-    "gpu_batches_per_second": 1.0490267598390537,
-    "producer_headroom": 3.5510005941231704,
+    "producer_batch_seconds": 0.12158256860925576,
+    "producer_p95_seconds": 0.3381743370046024,
+    "gpu_update_seconds": 0.954188086087161,
+    "producer_batches_per_second": 8.224863246752237,
+    "gpu_batches_per_second": 1.0480114084223162,
+    "producer_headroom": 7.84806651974715,
     "sustainable": true,
-    "buffer_seconds": 60.055665319410664,
+    "buffer_seconds": 60.11384942349114,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396663.21819072333,
-    "upload_bytes_per_second": 4792094.046324953,
+    "generation_bytes_per_second": 396155.61459481553,
+    "upload_bytes_per_second": 4831960.809389454,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.11469277659678,
+    "estimated_full_shards_per_hour": 85.00577286132192,
     "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": 1.0089386899999226,
-      "bytes": 4885086,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.4905629579989181,
       "bytes": 13725537,
@@ -562,6 +556,12 @@
       "bytes": 7296914,
       "pages": 42,
       "shards": 2
+    },
+    {
+      "seconds": 1.5529981709987624,
+      "bytes": 11067080,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -351,3 +351,4 @@
 {"step": 960100, "loss": 0.5036099553108215, "lr": 3e-05, "grad_norm": 0.3559221631435305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140532670170, "tokens_per_second": 140340.27813467168, "data_wait_s": 1.2641001376323402e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 960200, "loss": 0.5323472619056702, "lr": 3e-05, "grad_norm": 0.3651668859079228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140547305755, "tokens_per_second": 152614.19836806075, "data_wait_s": 1.561100361868739e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 960300, "loss": 0.4896109700202942, "lr": 3e-05, "grad_norm": 0.3447349619880113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140561809848, "tokens_per_second": 150716.92197168712, "data_wait_s": 1.5595003787893802e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 960400, "loss": 0.5473693013191223, "lr": 3e-05, "grad_norm": 0.3466411055769713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140576307698, "tokens_per_second": 150334.93396943517, "data_wait_s": 1.3853998098056763e-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:b6091ae3f602cc903387b5d31161545efb95aff0e09dd1fa8948f89c2734f8c1
-size 753300
+oid sha256:c5d88341c4f82552637f836297fe42724b9aeb4eeff50f992183d89b1c7c5686
+size 755418