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": 77313.27979805967,
-      "required_pages_per_second_upper_estimate": 0.005348712087954228
+      "required_token_positions_per_second": 77270.05087870843,
+      "required_pages_per_second_upper_estimate": 0.005345721411008597
     },
     {
       "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": 9664.159974757458,
-      "required_pages_per_second_upper_estimate": 0.0040461819708220255
+      "required_token_positions_per_second": 9658.756359838553,
+      "required_pages_per_second_upper_estimate": 0.004043919590095791
     },
     {
       "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": 19328.319949514917,
-      "required_pages_per_second_upper_estimate": 0.0040255793234095635
+      "required_token_positions_per_second": 19317.512719677106,
+      "required_pages_per_second_upper_estimate": 0.004023328462440164
     },
     {
       "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": 28992.479924272375,
-      "required_pages_per_second_upper_estimate": 0.0017380549025065401
+      "required_token_positions_per_second": 28976.26907951566,
+      "required_pages_per_second_upper_estimate": 0.0017370830871159017
     },
     {
       "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": 28992.479924272375,
-      "required_pages_per_second_upper_estimate": 0.0022540377896701886
+      "required_token_positions_per_second": 28976.26907951566,
+      "required_pages_per_second_upper_estimate": 0.0022527774677943246
     },
     {
       "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": 28992.479924272375,
-      "required_pages_per_second_upper_estimate": 0.37188440276897905
+      "required_token_positions_per_second": 28976.26907951566,
+      "required_pages_per_second_upper_estimate": 0.37167646745828886
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08804953088292677,
-    "producer_p95_seconds": 0.32013731899496634,
-    "gpu_update_seconds": 0.9536265980770143,
-    "producer_batches_per_second": 11.357243928188886,
-    "gpu_batches_per_second": 1.0486284694832313,
-    "producer_headroom": 10.830569890769596,
+    "producer_batch_seconds": 0.14516561347636525,
+    "producer_p95_seconds": 0.32808878499781713,
+    "gpu_update_seconds": 0.9541601068146258,
+    "producer_batches_per_second": 6.888683731996988,
+    "gpu_batches_per_second": 1.0480421397394264,
+    "producer_headroom": 6.572907205534421,
     "sustainable": true,
-    "buffer_seconds": 60.0784756788519,
+    "buffer_seconds": 60.112086729321426,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315897.32836375537,
+    "generation_bytes_per_second": 316359.30635941785,
     "upload_bytes_per_second": 4777316.618554914,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.78421295341964,
+    "estimated_full_shards_per_hour": 67.88334267699148,
     "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": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -369,3 +369,4 @@
 {"step": 796900, "loss": 0.48557382822036743, "lr": 3e-05, "grad_norm": 0.26081033136248116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116643322906, "tokens_per_second": 151187.62686386445, "data_wait_s": 1.1613999959081411e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 797000, "loss": 0.5508695244789124, "lr": 3e-05, "grad_norm": 0.2761141875675312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116657996194, "tokens_per_second": 152593.4678277973, "data_wait_s": 1.3636999938171357e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 797100, "loss": 0.5762922167778015, "lr": 3e-05, "grad_norm": 0.2785145102406768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116672709376, "tokens_per_second": 152721.36811185308, "data_wait_s": 1.0238996765110642e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 797200, "loss": 0.5406336784362793, "lr": 3e-05, "grad_norm": 0.26895965673834704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116687364548, "tokens_per_second": 151971.33816354838, "data_wait_s": 1.3086995750200003e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:102df1cf3290c4e40dc154ee1e38375c2eb5230240f5ca51ae39b8d3409533a3
-size 791424
+oid sha256:9cf133497caeeb04b899b1d158e25dd558aee38b78bf358ed5f53fd8e0830480
+size 793542