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": 76460.72729047562,
-      "required_pages_per_second_upper_estimate": 0.0052897305272852045
+      "required_token_positions_per_second": 77330.9309881231,
+      "required_pages_per_second_upper_estimate": 0.00534993323823922
     },
     {
       "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": 9557.590911309453,
-      "required_pages_per_second_upper_estimate": 0.004001563729371451
+      "required_token_positions_per_second": 9666.366373515388,
+      "required_pages_per_second_upper_estimate": 0.004047105743907125
     },
     {
       "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": 19115.181822618906,
-      "required_pages_per_second_upper_estimate": 0.003981188272407515
+      "required_token_positions_per_second": 19332.732747030775,
+      "required_pages_per_second_upper_estimate": 0.004026498392758821
     },
     {
       "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": 28672.77273392836,
-      "required_pages_per_second_upper_estimate": 0.0017188889446100304
+      "required_token_positions_per_second": 28999.099120546165,
+      "required_pages_per_second_upper_estimate": 0.0017384517132162263
     },
     {
       "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": 28672.77273392836,
-      "required_pages_per_second_upper_estimate": 0.0022291819618642556
+      "required_token_positions_per_second": 28999.099120546165,
+      "required_pages_per_second_upper_estimate": 0.0022545524030657084
     },
     {
       "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": 28672.77273392836,
-      "required_pages_per_second_upper_estimate": 0.367783542206082
+      "required_token_positions_per_second": 28999.099120546165,
+      "required_pages_per_second_upper_estimate": 0.37196930671163997
     }
   ],
   "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": 0.13091461985180786,
-    "producer_p95_seconds": 0.3275292789985542,
-    "gpu_update_seconds": 0.9642597266948044,
-    "producer_batches_per_second": 7.638566274202037,
-    "gpu_batches_per_second": 1.037064986036182,
-    "producer_headroom": 7.365561827902207,
+    "producer_batch_seconds": 0.12026974707811178,
+    "producer_p95_seconds": 0.3361881840028218,
+    "gpu_update_seconds": 0.9534089278108334,
+    "producer_batches_per_second": 8.314642911409205,
+    "gpu_batches_per_second": 1.04886787907068,
+    "producer_headroom": 7.927254783296596,
     "sustainable": true,
-    "buffer_seconds": 60.74836278177268,
+    "buffer_seconds": 60.064762452082505,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 281283.88039392483,
-    "upload_bytes_per_second": 4992195.9733101055,
+    "generation_bytes_per_second": 303878.09805717174,
+    "upload_bytes_per_second": 5010918.165214353,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.356972778924074,
+    "estimated_full_shards_per_hour": 65.20516592298856,
     "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": 1.1665725029961322,
-      "bytes": 1196873,
-      "pages": 37,
-      "shards": 1
-    },
     {
       "seconds": 1.1840190020011505,
       "bytes": 7418937,
@@ -562,6 +556,12 @@
       "bytes": 16175519,
       "pages": 45,
       "shards": 4
+    },
+    {
+      "seconds": 1.1545158149965573,
+      "bytes": 2752375,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -421,3 +421,4 @@
 {"step": 927100, "loss": 0.5215455889701843, "lr": 3e-05, "grad_norm": 0.33811687711354393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135701278418, "tokens_per_second": 153998.26457812803, "data_wait_s": 1.2737000361084938e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927200, "loss": 0.5180689692497253, "lr": 3e-05, "grad_norm": 0.3315590483140426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135715787627, "tokens_per_second": 151510.55934753237, "data_wait_s": 1.435800368199125e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 927300, "loss": 0.5017232894897461, "lr": 3e-05, "grad_norm": 0.3300251191732454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135730323364, "tokens_per_second": 151996.25186029857, "data_wait_s": 1.6120000509545207e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 927400, "loss": 0.5002279281616211, "lr": 3e-05, "grad_norm": 0.32860060915400224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135745010409, "tokens_per_second": 153240.5829542607, "data_wait_s": 1.6832003893796355e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0089844022279fe4718bb6e65d8ef1e08a982ccc8878b5caac0537621e88a96b
-size 902343
+oid sha256:770f66386be3136624d8b9c0b884c8cf7cbf803cf2cca843776c9bb2e035c335
+size 904461