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": 74880.88199754592,
-      "required_pages_per_second_upper_estimate": 0.006490485067686946
+      "required_token_positions_per_second": 74092.4167428623,
+      "required_pages_per_second_upper_estimate": 0.0064221428977578885
     },
     {
       "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": 9360.11024969324,
-      "required_pages_per_second_upper_estimate": 0.003733887924722052
+      "required_token_positions_per_second": 9261.552092857788,
+      "required_pages_per_second_upper_estimate": 0.003694571602384629
     },
     {
       "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": 18720.22049938648,
-      "required_pages_per_second_upper_estimate": 0.004330593547904887
+      "required_token_positions_per_second": 18523.104185715576,
+      "required_pages_per_second_upper_estimate": 0.004284994157865759
     },
     {
       "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": 28080.330749079716,
-      "required_pages_per_second_upper_estimate": 0.001709944816509476
+      "required_token_positions_per_second": 27784.656278573362,
+      "required_pages_per_second_upper_estimate": 0.0016919397925396961
     },
     {
       "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": 28080.330749079716,
-      "required_pages_per_second_upper_estimate": 0.0021617015923946965
+      "required_token_positions_per_second": 27784.656278573362,
+      "required_pages_per_second_upper_estimate": 0.002138939753175798
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28080.330749079716,
-      "required_pages_per_second_upper_estimate": 0.3486637290820333
+      "required_token_positions_per_second": 27784.656278573362,
+      "required_pages_per_second_upper_estimate": 0.3449924417171407
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.323573031273213,
-    "producer_p95_seconds": 0.3768727360002231,
-    "gpu_update_seconds": 0.9846037871511223,
-    "producer_batches_per_second": 3.090492418558941,
-    "gpu_batches_per_second": 1.0156369628573394,
-    "producer_headroom": 3.0429105394749647,
+    "producer_batch_seconds": 0.3677879314686834,
+    "producer_p95_seconds": 0.3837263660025201,
+    "gpu_update_seconds": 0.9950815918972247,
+    "producer_batches_per_second": 2.718958167024979,
+    "gpu_batches_per_second": 1.004942718409048,
+    "producer_headroom": 2.705585221145176,
     "sustainable": true,
-    "buffer_seconds": 60.06083101621846,
+    "buffer_seconds": 60.6999771057307,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420304.18954221735,
-    "upload_bytes_per_second": 5269719.956259195,
+    "generation_bytes_per_second": 392576.5087186024,
+    "upload_bytes_per_second": 5245253.394989282,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.18749489498035,
+    "estimated_full_shards_per_hour": 84.23778005760721,
     "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": 1.134857662000286,
-      "bytes": 5917855,
-      "pages": 41,
-      "shards": 2
-    },
     {
       "seconds": 1.5188524260011036,
       "bytes": 7655478,
@@ -562,6 +556,12 @@
       "bytes": 10463664,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.8019685460021719,
+      "bytes": 2186825,
+      "pages": 23,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -334,3 +334,4 @@
 {"step": 148400, "loss": 0.6222486458718777, "lr": 0.0003, "grad_norm": 0.1392784178267337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21717515881, "tokens_per_second": 150904.27421738647, "data_wait_s": 1.5240999346133322e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 148500, "loss": 0.6630819588899612, "lr": 0.0003, "grad_norm": 0.17121108387817255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21732122033, "tokens_per_second": 148464.6020109079, "data_wait_s": 1.3070006389170885e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 148600, "loss": 0.6224764958024025, "lr": 0.0003, "grad_norm": 0.14874271455697963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21746772551, "tokens_per_second": 150052.23278621814, "data_wait_s": 1.2166005035396665e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 148700, "loss": 0.6048832219094038, "lr": 0.0003, "grad_norm": 0.1411568040779498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21761408555, "tokens_per_second": 149694.91322863888, "data_wait_s": 1.347899524262175e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f74aaf3ef0dcdbb1c9fe0ac44f5dbd483f66a034f83ddbaf9621d8a5250b862
-size 716511
+oid sha256:7c6d0d3b24197caff4c313e9573b5d58e2119d81c4f8caa6039778e7af078785
+size 718629