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": 75299.42420610238,
-      "required_pages_per_second_upper_estimate": 0.0065267632455390905
+      "required_token_positions_per_second": 75065.15692667571,
+      "required_pages_per_second_upper_estimate": 0.006506457551503379
     },
     {
       "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": 9412.428025762798,
-      "required_pages_per_second_upper_estimate": 0.0037547582678166577
+      "required_token_positions_per_second": 9383.144615834464,
+      "required_pages_per_second_upper_estimate": 0.0037430766777702506
     },
     {
       "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": 18824.856051525596,
-      "required_pages_per_second_upper_estimate": 0.0043547991413694495
+      "required_token_positions_per_second": 18766.28923166893,
+      "required_pages_per_second_upper_estimate": 0.004341250738336435
     },
     {
       "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": 28237.284077288394,
-      "required_pages_per_second_upper_estimate": 0.0017195024507269125
+      "required_token_positions_per_second": 28149.433847503395,
+      "required_pages_per_second_upper_estimate": 0.0017141528326475404
     },
     {
       "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": 28237.284077288394,
-      "required_pages_per_second_upper_estimate": 0.0021737842940748207
+      "required_token_positions_per_second": 28149.433847503395,
+      "required_pages_per_second_upper_estimate": 0.002167021340200971
     },
     {
       "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": 28237.284077288394,
-      "required_pages_per_second_upper_estimate": 0.3506125641293864
+      "required_token_positions_per_second": 28149.433847503395,
+      "required_pages_per_second_upper_estimate": 0.3495217582912623
     }
   ],
   "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": 0.09216576511721541,
-    "producer_p95_seconds": 0.27992409199941903,
-    "gpu_update_seconds": 0.9791309930630913,
-    "producer_batches_per_second": 10.850015716011372,
-    "gpu_batches_per_second": 1.0213138048787758,
-    "producer_headroom": 10.623586662768362,
+    "producer_batch_seconds": 0.1598785053984102,
+    "producer_p95_seconds": 0.23976872600178467,
+    "gpu_update_seconds": 0.9821867164284774,
+    "producer_batches_per_second": 6.254749489357835,
+    "gpu_batches_per_second": 1.0181363515445383,
+    "producer_headroom": 6.143331863035067,
     "sustainable": true,
-    "buffer_seconds": 60.70612156991166,
+    "buffer_seconds": 60.895576418565604,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 266204.438409972,
-    "upload_bytes_per_second": 5336203.8496810775,
+    "generation_bytes_per_second": 287220.3718771834,
+    "upload_bytes_per_second": 5394101.357329655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.12127556061144,
+    "estimated_full_shards_per_hour": 61.6308056567824,
     "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": 0.8817397439997876,
-      "bytes": 1063209,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.6997340180005267,
       "bytes": 11573294,
@@ -562,6 +556,12 @@
       "bytes": 11625452,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.1623286959984398,
+      "bytes": 7130450,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -195,3 +195,4 @@
 {"step": 214500, "loss": 0.6368815768510103, "lr": 0.0003, "grad_norm": 0.14979569600535872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31395891452, "tokens_per_second": 149599.7179484442, "data_wait_s": 1.168600283563137e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 214600, "loss": 0.6312737930566072, "lr": 0.0003, "grad_norm": 0.14112733343290895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31410633289, "tokens_per_second": 149667.90267675795, "data_wait_s": 1.8078000721288845e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 214700, "loss": 0.5823081564158201, "lr": 0.0003, "grad_norm": 0.14617640280384095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31425411172, "tokens_per_second": 150158.34901225442, "data_wait_s": 1.4105000445852056e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 214800, "loss": 0.5876647941768169, "lr": 0.0003, "grad_norm": 0.14906130716057017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31440150645, "tokens_per_second": 150202.25105826938, "data_wait_s": 1.3276003301143646e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5de87dab1d87eb94e0da52f4fcfe9ba478bf73cb842854fd2590b8d381edda7
-size 419760
+oid sha256:1cb1948fcdf3d2b8bd885dd236d725c358c67f9d5dac7c246c7d6dd3a7059c7f
+size 421878