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": 76892.60573705634,
-      "required_pages_per_second_upper_estimate": 0.005319608880315716
+      "required_token_positions_per_second": 77414.22351264503,
+      "required_pages_per_second_upper_estimate": 0.00535569560835093
     },
     {
       "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": 9611.575717132042,
-      "required_pages_per_second_upper_estimate": 0.00402416604023843
+      "required_token_positions_per_second": 9676.777939080628,
+      "required_pages_per_second_upper_estimate": 0.004051464848991079
     },
     {
       "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": 19223.151434264084,
-      "required_pages_per_second_upper_estimate": 0.004003675495163071
+      "required_token_positions_per_second": 19353.555878161256,
+      "required_pages_per_second_upper_estimate": 0.0040308353018304035
     },
     {
       "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": 28834.727151396124,
-      "required_pages_per_second_upper_estimate": 0.0017285978646471459
+      "required_token_positions_per_second": 29030.33381724188,
+      "required_pages_per_second_upper_estimate": 0.001740324185590514
     },
     {
       "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": 28834.727151396124,
-      "required_pages_per_second_upper_estimate": 0.0022417732054601765
+      "required_token_positions_per_second": 29030.33381724188,
+      "required_pages_per_second_upper_estimate": 0.002256980763346889
     },
     {
       "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": 28834.727151396124,
-      "required_pages_per_second_upper_estimate": 0.36986091958025324
+      "required_token_positions_per_second": 29030.33381724188,
+      "required_pages_per_second_upper_estimate": 0.37236995186364824
     }
   ],
   "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": 0.42652256832792546,
-    "producer_p95_seconds": 0.35901466500217794,
-    "gpu_update_seconds": 0.9588438224102056,
-    "producer_batches_per_second": 2.344541823238683,
-    "gpu_batches_per_second": 1.0429227123624176,
-    "producer_headroom": 2.248049443594771,
+    "producer_batch_seconds": 0.11551966793717838,
+    "producer_p95_seconds": 0.3608287859970005,
+    "gpu_update_seconds": 0.9523831235994649,
+    "producer_batches_per_second": 8.656534578543088,
+    "gpu_batches_per_second": 1.0499976062370473,
+    "producer_headroom": 8.244337441459644,
     "sustainable": true,
-    "buffer_seconds": 60.40716081184296,
+    "buffer_seconds": 60.000136786766284,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325570.10278753965,
-    "upload_bytes_per_second": 4820079.831487125,
+    "generation_bytes_per_second": 325549.16798080783,
+    "upload_bytes_per_second": 4881245.2461191835,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.85976517409937,
+    "estimated_full_shards_per_hour": 69.85527305191208,
     "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": 1.3873778990018764,
-      "bytes": 7922754,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.5672195150000334,
       "bytes": 991660,
@@ -562,6 +556,12 @@
       "bytes": 11019556,
       "pages": 29,
       "shards": 4
+    },
+    {
+      "seconds": 1.1776043950012536,
+      "bytes": 12312597,
+      "pages": 48,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -372,3 +372,4 @@
 {"step": 962200, "loss": 0.5208715200424194, "lr": 3e-05, "grad_norm": 0.33088779700782583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140840430506, "tokens_per_second": 152777.79002305097, "data_wait_s": 1.3284996384754777e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 962300, "loss": 0.5374898314476013, "lr": 3e-05, "grad_norm": 0.35127594183016225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140855028084, "tokens_per_second": 146374.7713553694, "data_wait_s": 1.471900031901896e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 962400, "loss": 0.5576936602592468, "lr": 3e-05, "grad_norm": 0.3441921049931057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140869619198, "tokens_per_second": 151438.39169458006, "data_wait_s": 1.3286000466905534e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 962500, "loss": 0.5486325025558472, "lr": 3e-05, "grad_norm": 0.3615660195929653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140884058809, "tokens_per_second": 150130.3312679302, "data_wait_s": 1.655399682931602e-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 sha256:1f481aeff93fa367bd69cf40ad19326100e3c189d88232e310a417cbcaecd195
-size 797778
+oid sha256:e0ecfbb209b1ba944a924b427f46d364bcd82dda26103242eb2741a8e2130208
+size 799896