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": 77170.39332101293,
-      "required_pages_per_second_upper_estimate": 0.005338826869929831
+      "required_token_positions_per_second": 76877.0100694814,
+      "required_pages_per_second_upper_estimate": 0.005318529935845947
     },
     {
       "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": 9646.299165126617,
-      "required_pages_per_second_upper_estimate": 0.004038704022805709
+      "required_token_positions_per_second": 9609.626258685175,
+      "required_pages_per_second_upper_estimate": 0.004023349842696049
     },
     {
       "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": 19292.598330253233,
-      "required_pages_per_second_upper_estimate": 0.004018139452159805
+      "required_token_positions_per_second": 19219.25251737035,
+      "required_pages_per_second_upper_estimate": 0.004002863453595459
     },
     {
       "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": 28938.89749537985,
-      "required_pages_per_second_upper_estimate": 0.0017348427177100649
+      "required_token_positions_per_second": 28828.878776055528,
+      "required_pages_per_second_upper_estimate": 0.0017282472634754292
     },
     {
       "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": 28938.89749537985,
-      "required_pages_per_second_upper_estimate": 0.0022498719915079917
+      "required_token_positions_per_second": 28828.878776055528,
+      "required_pages_per_second_upper_estimate": 0.0022413185200017296
     },
     {
       "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": 28938.89749537985,
-      "required_pages_per_second_upper_estimate": 0.37119710490347546
+      "required_token_positions_per_second": 28828.878776055528,
+      "required_pages_per_second_upper_estimate": 0.36978590290088026
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1137480395858006,
-    "producer_p95_seconds": 0.34392818799824454,
-    "gpu_update_seconds": 0.9553923056126812,
-    "producer_batches_per_second": 8.79136030512153,
-    "gpu_batches_per_second": 1.0466904476048846,
-    "producer_headroom": 8.399197991381865,
+    "producer_batch_seconds": 0.12322914650047778,
+    "producer_p95_seconds": 0.33006875999853946,
+    "gpu_update_seconds": 0.9590383384234723,
+    "producer_batches_per_second": 8.114963289112149,
+    "gpu_batches_per_second": 1.0427111825830269,
+    "producer_headroom": 7.782560909157591,
     "sustainable": true,
-    "buffer_seconds": 60.18971525359891,
+    "buffer_seconds": 60.41941532067875,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331531.55085523945,
-    "upload_bytes_per_second": 4593959.83944083,
+    "generation_bytes_per_second": 332195.9061095729,
+    "upload_bytes_per_second": 4549082.312303949,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.13895315401923,
+    "estimated_full_shards_per_hour": 71.2815083262004,
     "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": 1.108235388997855,
-      "bytes": 5901441,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0388182249989768,
       "bytes": 6175327,
@@ -562,6 +556,12 @@
       "bytes": 6133990,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.2553032559953863,
+      "bytes": 2375236,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -398,3 +398,4 @@
 {"step": 799800, "loss": 0.5266774892807007, "lr": 3e-05, "grad_norm": 0.27029494828643785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117067352741, "tokens_per_second": 150052.93743880262, "data_wait_s": 1.3017001037951559e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 799900, "loss": 0.5137812495231628, "lr": 3e-05, "grad_norm": 0.2582192352107411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117081959453, "tokens_per_second": 151264.61480310414, "data_wait_s": 1.623500429559499e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800000, "loss": 0.5484288930892944, "lr": 3e-05, "grad_norm": 0.2769821685590719, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117096573705, "tokens_per_second": 151859.07763975218, "data_wait_s": 1.7142003343906254e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 800100, "loss": 0.5328884124755859, "lr": 3e-05, "grad_norm": 0.26596277544685065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117111234819, "tokens_per_second": 144089.23081662506, "data_wait_s": 1.2935997801832855e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ddaab7a3ae74eb714c8428f9bfb2f2548e38f9b28517d6e9ea8bd753b961608e
-size 853629
+oid sha256:a946915a861a6aa241a740f9c11922dce1b49b08172c59cb07b6eb12e7728f6c
+size 855747