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": 76749.58970179378,
-      "required_pages_per_second_upper_estimate": 0.005309714699153332
+      "required_token_positions_per_second": 76304.66383202442,
+      "required_pages_per_second_upper_estimate": 0.005278933695112437
     },
     {
       "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": 9593.698712724223,
-      "required_pages_per_second_upper_estimate": 0.004016681311807179
+      "required_token_positions_per_second": 9538.082979003053,
+      "required_pages_per_second_upper_estimate": 0.003993396165486712
     },
     {
       "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": 19187.397425448446,
-      "required_pages_per_second_upper_estimate": 0.003996228878023393
+      "required_token_positions_per_second": 19076.165958006106,
+      "required_pages_per_second_upper_estimate": 0.003973062296726211
     },
     {
       "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": 28781.096138172667,
-      "required_pages_per_second_upper_estimate": 0.0017253827672942153
+      "required_token_positions_per_second": 28614.24893700916,
+      "required_pages_per_second_upper_estimate": 0.0017153805323453881
     },
     {
       "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": 28781.096138172667,
-      "required_pages_per_second_upper_estimate": 0.0022376036300799495
+      "required_token_positions_per_second": 28614.24893700916,
+      "required_pages_per_second_upper_estimate": 0.0022246319940727654
     },
     {
       "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": 28781.096138172667,
-      "required_pages_per_second_upper_estimate": 0.36917299852711827
+      "required_token_positions_per_second": 28614.24893700916,
+      "required_pages_per_second_upper_estimate": 0.3670328617771599
     }
   ],
   "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": 0.08985940417187521,
-    "producer_p95_seconds": 0.26756889000171213,
-    "gpu_update_seconds": 0.960630542605713,
-    "producer_batches_per_second": 11.128495778664272,
-    "gpu_batches_per_second": 1.0409829332383054,
-    "producer_headroom": 10.690372938243646,
+    "producer_batch_seconds": 0.11807231580462485,
+    "producer_p95_seconds": 0.26063793399953283,
+    "gpu_update_seconds": 0.9662318958943762,
+    "producer_batches_per_second": 8.469385843627455,
+    "gpu_batches_per_second": 1.0349482399091856,
+    "producer_headroom": 8.183390740749147,
     "sustainable": true,
-    "buffer_seconds": 60.51972418415992,
+    "buffer_seconds": 60.8726094413457,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298336.73584698414,
-    "upload_bytes_per_second": 1890847.884392522,
+    "generation_bytes_per_second": 330713.6353892994,
+    "upload_bytes_per_second": 1865121.6296303656,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.01611858899253,
+    "estimated_full_shards_per_hour": 70.96344753512608,
     "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": 1.2059528549998504,
-      "bytes": 7149120,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.9931068189998769,
       "bytes": 6968332,
@@ -562,6 +556,12 @@
       "bytes": 5596025,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.1274023559999478,
+      "bytes": 1023034,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -211,3 +211,4 @@
 {"step": 666100, "loss": 0.579171895980835, "lr": 0.0001, "grad_norm": 0.1892237160171339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97500725009, "tokens_per_second": 150352.89336690045, "data_wait_s": 1.7468002624809742e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 666200, "loss": 0.5548118948936462, "lr": 0.0001, "grad_norm": 0.18941556656407524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97515163744, "tokens_per_second": 150559.8314973522, "data_wait_s": 1.882100332295522e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 666300, "loss": 0.5639391541481018, "lr": 0.0001, "grad_norm": 0.20244867513736692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97529618966, "tokens_per_second": 150769.23351798995, "data_wait_s": 1.2085998605471104e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 666400, "loss": 0.615656316280365, "lr": 0.0001, "grad_norm": 0.19269532786653254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97544316740, "tokens_per_second": 153093.79737789245, "data_wait_s": 1.0648000170476735e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e16efdd9098ca0f3d3722883ee85bbb3ecbeed9d7c1fb82bc5b322992c15e16
-size 454431
+oid sha256:337e9c1f9828e03bc43647bef3282be8c789f1882da01a3a2b6405ffb3693f51
+size 456549