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": 76813.07307891943,
-      "required_pages_per_second_upper_estimate": 0.005314106626484613
+      "required_token_positions_per_second": 76445.51233252675,
+      "required_pages_per_second_upper_estimate": 0.005288677921190735
     },
     {
       "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": 9601.63413486493,
-      "required_pages_per_second_upper_estimate": 0.004020003707347035
+      "required_token_positions_per_second": 9555.689041565844,
+      "required_pages_per_second_upper_estimate": 0.004000767456225358
     },
     {
       "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": 19203.26826972986,
-      "required_pages_per_second_upper_estimate": 0.003999534356344901
+      "required_token_positions_per_second": 19111.378083131687,
+      "required_pages_per_second_upper_estimate": 0.003980396053783684
     },
     {
       "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": 28804.902404594788,
-      "required_pages_per_second_upper_estimate": 0.0017268099166161573
+      "required_token_positions_per_second": 28667.067124697533,
+      "required_pages_per_second_upper_estimate": 0.0017185469020486032
     },
     {
       "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": 28804.902404594788,
-      "required_pages_per_second_upper_estimate": 0.00223945446258157
+      "required_token_positions_per_second": 28667.067124697533,
+      "required_pages_per_second_upper_estimate": 0.0022287383758428816
     },
     {
       "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": 28804.902404594788,
-      "required_pages_per_second_upper_estimate": 0.36947835975160387
+      "required_token_positions_per_second": 28667.067124697533,
+      "required_pages_per_second_upper_estimate": 0.36771035677707486
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09489528160196414,
-    "producer_p95_seconds": 0.33863073700194946,
-    "gpu_update_seconds": 0.9598366143254057,
-    "producer_batches_per_second": 10.537931740320607,
-    "gpu_batches_per_second": 1.0418439816476703,
-    "producer_headroom": 10.11469272362156,
+    "producer_batch_seconds": 0.3456384176249685,
+    "producer_p95_seconds": 0.3505993880025926,
+    "gpu_update_seconds": 0.9644516434044424,
+    "producer_batches_per_second": 2.893196904648024,
+    "gpu_batches_per_second": 1.0368586199615717,
+    "producer_headroom": 2.790348509380433,
     "sustainable": true,
-    "buffer_seconds": 60.46970670250056,
+    "buffer_seconds": 60.76045353447987,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393698.5482561809,
-    "upload_bytes_per_second": 5174347.025328171,
+    "generation_bytes_per_second": 363923.12344089994,
+    "upload_bytes_per_second": 5049410.329721906,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.47854362262792,
+    "estimated_full_shards_per_hour": 78.0894305936837,
     "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": 1.1330570440004522,
-      "bytes": 12010212,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.1789866869985417,
       "bytes": 2676190,
@@ -562,6 +556,12 @@
       "bytes": 7398571,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.022340563002217,
+      "bytes": 921929,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -394,3 +394,4 @@
 {"step": 436500, "loss": 0.5513224005699158, "lr": 0.0003, "grad_norm": 0.15447092581467953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63890921532, "tokens_per_second": 152774.02123816113, "data_wait_s": 1.5495999832637608e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 436600, "loss": 0.6376541256904602, "lr": 0.0003, "grad_norm": 0.1931321346204208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63905645007, "tokens_per_second": 151648.38823333982, "data_wait_s": 9.438001143280417e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 436700, "loss": 0.5688666701316833, "lr": 0.0003, "grad_norm": 0.14958992454250414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63920360215, "tokens_per_second": 151443.07067067074, "data_wait_s": 9.973002306651324e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 436800, "loss": 0.6163980960845947, "lr": 0.0003, "grad_norm": 0.1846098944574017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63934942003, "tokens_per_second": 150709.02231486028, "data_wait_s": 1.6042999050114304e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a6be0e6c851efb41aaeba4c81c7f66a147cbeba2e3eef1703125434d6627860
-size 845157
+oid sha256:2e305e0bb053135f4f2dfa0b49595e221a685f74812ce847b2a3ee12128a7325
+size 847275