Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75177.80113748496,
-      "required_pages_per_second_upper_estimate": 0.006516221266202182
+      "required_token_positions_per_second": 75508.17847187209,
+      "required_pages_per_second_upper_estimate": 0.006544857536213159
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9397.22514218562,
-      "required_pages_per_second_upper_estimate": 0.003748693610254356
+      "required_token_positions_per_second": 9438.522308984011,
+      "required_pages_per_second_upper_estimate": 0.0037651676675379015
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18794.45028437124,
-      "required_pages_per_second_upper_estimate": 0.004347765302261514
+      "required_token_positions_per_second": 18877.044617968022,
+      "required_pages_per_second_upper_estimate": 0.004366872047728506
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28191.67542655686,
-      "required_pages_per_second_upper_estimate": 0.001716725123187459
+      "required_token_positions_per_second": 28315.566926952033,
+      "required_pages_per_second_upper_estimate": 0.001724269465553058
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28191.67542655686,
-      "required_pages_per_second_upper_estimate": 0.0021702732138886814
+      "required_token_positions_per_second": 28315.566926952033,
+      "required_pages_per_second_upper_estimate": 0.0021798107245427248
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28191.67542655686,
-      "required_pages_per_second_upper_estimate": 0.35004625732963557
+      "required_token_positions_per_second": 28315.566926952033,
+      "required_pages_per_second_upper_estimate": 0.3515845751263647
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12101144449999524,
-    "producer_p95_seconds": 0.14369549399998505,
-    "gpu_update_seconds": 0.9807150366790648,
-    "producer_batches_per_second": 8.26368120909456,
-    "gpu_batches_per_second": 1.0196641864350715,
-    "producer_headroom": 8.10431642008127,
+    "producer_batch_seconds": 0.050682581375003366,
+    "producer_p95_seconds": 0.13704574799999136,
+    "gpu_update_seconds": 0.9764240310400915,
+    "producer_batches_per_second": 19.730644589724857,
+    "gpu_batches_per_second": 1.0241452158185775,
+    "producer_headroom": 19.265475525318518,
     "sustainable": true,
-    "buffer_seconds": 60.804332274102016,
+    "buffer_seconds": 60.53828992448567,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 131,
+    "upload_pages_per_commit": 143,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 778617.9773980809,
-    "upload_bytes_per_second": 4417317.494560352,
+    "generation_bytes_per_second": 756727.9281404478,
+    "upload_bytes_per_second": 4873718.004538263,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.53664632538232,
+    "estimated_full_shards_per_hour": 81.18809882717169,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 3640398,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.3395652879999034,
+      "bytes": 11637618,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -6,3 +6,4 @@
 {"step": 600, "loss": 1.4354848563671112, "lr": 0.00017999999999999998, "grad_norm": 1.4426431949597145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 88325921, "tokens_per_second": 150330.61897931347, "data_wait_s": 9.991000069931033e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 700, "loss": 1.4169969446957111, "lr": 0.00020999999999999998, "grad_norm": 1.1299403160160901, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 103081101, "tokens_per_second": 149879.6582243547, "data_wait_s": 1.00040001598245e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 800, "loss": 1.3345604501664639, "lr": 0.00023999999999999998, "grad_norm": 0.8847908940671446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 117723562, "tokens_per_second": 148830.22167264167, "data_wait_s": 1.11660001493874e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 900, "loss": 1.2485260888934135, "lr": 0.00026999999999999995, "grad_norm": 0.5126180461192602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 132386398, "tokens_per_second": 148764.82655347895, "data_wait_s": 2.0052000081705046e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2aa8367b1eb35fd9339f5c1956b49ce671c0ffff560282b983bdd5a15187f4e
-size 16790
+oid sha256:a5c3bb1cd0e2b3732a29675af263bcb112e02bb38019c04641ef3124749f57ce
+size 18873