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": 75184.60877564199,
-      "required_pages_per_second_upper_estimate": 0.0065168113350238316
+      "required_token_positions_per_second": 74120.32407300951,
+      "required_pages_per_second_upper_estimate": 0.006424561834404567
     },
     {
       "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": 9398.076096955248,
-      "required_pages_per_second_upper_estimate": 0.0037490330688348685
+      "required_token_positions_per_second": 9265.04050912619,
+      "required_pages_per_second_upper_estimate": 0.0036959631837905653
     },
     {
       "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": 18796.152193910497,
-      "required_pages_per_second_upper_estimate": 0.004348159009080837
+      "required_token_positions_per_second": 18530.08101825238,
+      "required_pages_per_second_upper_estimate": 0.004286608125285089
     },
     {
       "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": 28194.228290865747,
-      "required_pages_per_second_upper_estimate": 0.0017168805792300264
+      "required_token_positions_per_second": 27795.121527378564,
+      "required_pages_per_second_upper_estimate": 0.0016925770712850159
     },
     {
       "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": 28194.228290865747,
-      "required_pages_per_second_upper_estimate": 0.0021704697404499604
+      "required_token_positions_per_second": 27795.121527378564,
+      "required_pages_per_second_upper_estimate": 0.002139745396998485
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28194.228290865747,
-      "required_pages_per_second_upper_estimate": 0.3500779553604647
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27795.121527378564,
+      "required_pages_per_second_upper_estimate": 0.34512238508236665
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0508377221562597,
-    "producer_p95_seconds": 0.17404284399981407,
-    "gpu_update_seconds": 0.9806262372131422,
-    "producer_batches_per_second": 19.67043285154091,
-    "gpu_batches_per_second": 1.0197565209369845,
-    "producer_headroom": 19.289342551560342,
+    "producer_batch_seconds": 0.2458600116328178,
+    "producer_p95_seconds": 0.17853448999994725,
+    "gpu_update_seconds": 0.9947069298749549,
+    "producer_batches_per_second": 4.067355213069218,
+    "gpu_batches_per_second": 1.0053212357992827,
+    "producer_headroom": 4.045826416702975,
     "sustainable": true,
-    "buffer_seconds": 60.79882670721482,
+    "buffer_seconds": 60.677122722372246,
     "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": 361660.6216724265,
-    "upload_bytes_per_second": 4802615.754091308,
+    "generation_bytes_per_second": 382337.89580150624,
+    "upload_bytes_per_second": 4905767.179069497,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.60395038251492,
+    "estimated_full_shards_per_hour": 82.04081206830874,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 1218836,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1421018979999644,
+      "bytes": 6132811,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -33,3 +33,4 @@
 {"step": 198300, "loss": 0.5880254246294498, "lr": 0.0003, "grad_norm": 0.13477129061986676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29023536881, "tokens_per_second": 148109.66909596414, "data_wait_s": 1.2978000086150132e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 198400, "loss": 0.6215808764100075, "lr": 0.0003, "grad_norm": 0.1683647299050308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29038047497, "tokens_per_second": 146338.90368490835, "data_wait_s": 1.6541999684704933e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 198500, "loss": 0.5699267964810133, "lr": 0.0003, "grad_norm": 0.13448268142187705, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29052619260, "tokens_per_second": 146686.85576588917, "data_wait_s": 1.6292000054818345e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 198600, "loss": 0.6101530659943819, "lr": 0.0003, "grad_norm": 0.14258927592460816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29067243499, "tokens_per_second": 148175.9281633369, "data_wait_s": 1.7857999409898184e-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 sha256:147702fc095c883ebdc67b3b7d7d0f5123a0f9d0df3c8522417e7fdc220b9d46
-size 74295
+oid sha256:fd959e9a38184ec42f704b8599ec23746d98910374d31b6423eaa0cae628b814
+size 76413