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": 74965.35333864111,
-      "required_pages_per_second_upper_estimate": 0.006497806829442421
+      "required_token_positions_per_second": 75648.12362092586,
+      "required_pages_per_second_upper_estimate": 0.006556987627045403
     },
     {
       "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": 9370.66916733014,
-      "required_pages_per_second_upper_estimate": 0.0037381000348372983
+      "required_token_positions_per_second": 9456.015452615733,
+      "required_pages_per_second_upper_estimate": 0.0037721459440784
     },
     {
       "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": 18741.33833466028,
-      "required_pages_per_second_upper_estimate": 0.004335478787434273
+      "required_token_positions_per_second": 18912.030905231466,
+      "required_pages_per_second_upper_estimate": 0.004374965509549278
     },
     {
       "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": 28112.007501990418,
-      "required_pages_per_second_upper_estimate": 0.0017118737645666594
+      "required_token_positions_per_second": 28368.0463578472,
+      "required_pages_per_second_upper_estimate": 0.001727465187556279
     },
     {
       "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": 28112.007501990418,
-      "required_pages_per_second_upper_estimate": 0.0021641401565206265
+      "required_token_positions_per_second": 28368.0463578472,
+      "required_pages_per_second_upper_estimate": 0.0021838507364053974
     },
     {
       "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": 28112.007501990418,
-      "required_pages_per_second_upper_estimate": 0.349057048337912
+      "required_token_positions_per_second": 28368.0463578472,
+      "required_pages_per_second_upper_estimate": 0.35223619402072587
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08917208794539988,
-    "producer_p95_seconds": 0.348272448998614,
-    "gpu_update_seconds": 0.9834943306002759,
-    "producer_batches_per_second": 11.214271450190788,
-    "gpu_batches_per_second": 1.01678267874676,
-    "producer_headroom": 11.029172393075175,
+    "producer_batch_seconds": 0.10788504744539296,
+    "producer_p95_seconds": 0.35009541399995214,
+    "gpu_update_seconds": 0.9746176966589728,
+    "producer_batches_per_second": 9.269125088962486,
+    "gpu_batches_per_second": 1.0260433433827836,
+    "producer_headroom": 9.033853344248515,
     "sustainable": true,
-    "buffer_seconds": 60.976648497217106,
+    "buffer_seconds": 60.42629719285631,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332448.4542040346,
-    "upload_bytes_per_second": 3707334.2140109786,
+    "generation_bytes_per_second": 371770.9009268763,
+    "upload_bytes_per_second": 3680294.027406763,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.33569926825312,
+    "estimated_full_shards_per_hour": 79.77338095526424,
     "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": 1.5110167620005086,
-      "bytes": 10255842,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 8.627843940001185,
       "bytes": 8388512,
@@ -562,6 +556,12 @@
       "bytes": 8396980,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1415953719988465,
+      "bytes": 5808415,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -320,3 +320,4 @@
 {"step": 72000, "loss": 0.6382339969277382, "lr": 0.0003, "grad_norm": 0.13505580527927427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10537417623, "tokens_per_second": 149394.73376919742, "data_wait_s": 8.438997610937804e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 72100, "loss": 0.605940455570817, "lr": 0.0003, "grad_norm": 0.14389794579566484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10552078089, "tokens_per_second": 149666.46635261903, "data_wait_s": 8.277002052636817e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 72200, "loss": 0.6321951113641262, "lr": 0.0003, "grad_norm": 0.1372418156646615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10566890635, "tokens_per_second": 152371.74784975115, "data_wait_s": 1.0890998964896426e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 72300, "loss": 0.6349485367536545, "lr": 0.0003, "grad_norm": 0.12504722232350515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10581358498, "tokens_per_second": 147930.35049789556, "data_wait_s": 8.243998308898881e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2cd61019312b6eb048abbef758ae85883a7af6fb20051c473bb5e0f6fb9d4a3
-size 686859
+oid sha256:c2dc85f0c23dc80d8a48736726f2e1d5775e7ddb6f659a8297ef2f9fb5b4ad4d
+size 688977