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": 74682.33390672976,
-      "required_pages_per_second_upper_estimate": 0.006473275422390537
+      "required_token_positions_per_second": 74268.41210219522,
+      "required_pages_per_second_upper_estimate": 0.006437397729448705
     },
     {
       "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": 9335.29173834122,
-      "required_pages_per_second_upper_estimate": 0.0037239874494739192
+      "required_token_positions_per_second": 9283.551512774402,
+      "required_pages_per_second_upper_estimate": 0.0037033474999100054
     },
     {
       "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": 18670.58347668244,
-      "required_pages_per_second_upper_estimate": 0.0043191108962840705
+      "required_token_positions_per_second": 18567.103025548804,
+      "required_pages_per_second_upper_estimate": 0.004295172515108042
     },
     {
       "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": 28005.87521502366,
-      "required_pages_per_second_upper_estimate": 0.0017054108651234593
+      "required_token_positions_per_second": 27850.654538323208,
+      "required_pages_per_second_upper_estimate": 0.0016959587402923537
     },
     {
       "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": 28005.87521502366,
-      "required_pages_per_second_upper_estimate": 0.0021559697992769517
+      "required_token_positions_per_second": 27850.654538323208,
+      "required_pages_per_second_upper_estimate": 0.002144020481906217
     },
     {
       "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": 28005.87521502366,
-      "required_pages_per_second_upper_estimate": 0.3477392405356999
+      "required_token_positions_per_second": 27850.654538323208,
+      "required_pages_per_second_upper_estimate": 0.34581191922126736
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11516989053906457,
-    "producer_p95_seconds": 0.21251288300027227,
-    "gpu_update_seconds": 0.9872214236378629,
-    "producer_batches_per_second": 8.682824958150057,
-    "gpu_batches_per_second": 1.012943982024872,
-    "producer_headroom": 8.571870816383267,
+    "producer_batch_seconds": 0.3851833753359273,
+    "producer_p95_seconds": 0.20662466700014193,
+    "gpu_update_seconds": 0.9927235269086998,
+    "producer_batches_per_second": 2.5961660446219335,
+    "gpu_batches_per_second": 1.0073298082437503,
+    "producer_headroom": 2.577275112257695,
     "sustainable": true,
-    "buffer_seconds": 60.22050684190964,
+    "buffer_seconds": 60.55613514143069,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326779.8097270122,
-    "upload_bytes_per_second": 5226382.210056248,
+    "generation_bytes_per_second": 288705.8574664902,
+    "upload_bytes_per_second": 5268964.064870502,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.11934012277388,
+    "estimated_full_shards_per_hour": 61.94955628391293,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 2561145,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.5218778089993066,
+      "bytes": 10001472,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -74,3 +74,4 @@
 {"step": 202400, "loss": 0.6051944028586149, "lr": 0.0003, "grad_norm": 0.15517594246542166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29623586994, "tokens_per_second": 148461.85263258754, "data_wait_s": 2.168100036215037e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 202500, "loss": 0.6498415544629097, "lr": 0.0003, "grad_norm": 0.15196157946285538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29638299561, "tokens_per_second": 148997.07727475307, "data_wait_s": 2.0152000615780707e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 202600, "loss": 0.6185240875929594, "lr": 0.0003, "grad_norm": 0.15168966879030246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29652808878, "tokens_per_second": 145896.18207990756, "data_wait_s": 8.65400033944752e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 202700, "loss": 0.6121512372046709, "lr": 0.0003, "grad_norm": 0.1389628787426531, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29667447656, "tokens_per_second": 148604.21248002467, "data_wait_s": 1.978900036192499e-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:aec69e8584e7bbc21e7141f1035b569b3575c85f746ca7bf1f7c966a77332ff0
-size 161916
+oid sha256:22ed88c3b95df76e15b03211d864cc442105c8881c4c5985c00fb65700f24711
+size 164034