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": 75336.3586036016,
-      "required_pages_per_second_upper_estimate": 0.006529964625504949
+      "required_token_positions_per_second": 75673.99018023415,
+      "required_pages_per_second_upper_estimate": 0.00655922967484805
     },
     {
       "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": 9417.0448254502,
-      "required_pages_per_second_upper_estimate": 0.0037565999782392693
+      "required_token_positions_per_second": 9459.248772529269,
+      "required_pages_per_second_upper_estimate": 0.00377343576373435
     },
     {
       "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": 18834.0896509004,
-      "required_pages_per_second_upper_estimate": 0.004356935172079013
+      "required_token_positions_per_second": 18918.497545058537,
+      "required_pages_per_second_upper_estimate": 0.004376461452864297
     },
     {
       "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": 28251.134476350602,
-      "required_pages_per_second_upper_estimate": 0.0017203458673623732
+      "required_token_positions_per_second": 28377.746317587804,
+      "required_pages_per_second_upper_estimate": 0.0017280558642127262
     },
     {
       "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": 28251.134476350602,
-      "required_pages_per_second_upper_estimate": 0.002174850536134988
+      "required_token_positions_per_second": 28377.746317587804,
+      "required_pages_per_second_upper_estimate": 0.0021845974661574364
     },
     {
       "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": 28251.134476350602,
-      "required_pages_per_second_upper_estimate": 0.35078453973143525
+      "required_token_positions_per_second": 28377.746317587804,
+      "required_pages_per_second_upper_estimate": 0.35235663505702725
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.45254392154691914,
-    "producer_p95_seconds": 0.31734839099954115,
-    "gpu_update_seconds": 0.9786509643761211,
-    "producer_batches_per_second": 2.2097302656982465,
-    "gpu_batches_per_second": 1.021814759705968,
-    "producer_headroom": 2.1625546555366912,
+    "producer_batch_seconds": 0.13011241470323398,
+    "producer_p95_seconds": 0.3252753359993221,
+    "gpu_update_seconds": 0.9742845570109446,
+    "producer_batches_per_second": 7.6856616817145635,
+    "gpu_batches_per_second": 1.0263941810470125,
+    "producer_headroom": 7.488021486905265,
     "sustainable": true,
-    "buffer_seconds": 60.676359791319506,
+    "buffer_seconds": 60.40564253467857,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347506.8508323707,
-    "upload_bytes_per_second": 4032104.024444273,
+    "generation_bytes_per_second": 348921.1610620921,
+    "upload_bytes_per_second": 4134405.775571858,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.56688064316121,
+    "estimated_full_shards_per_hour": 74.8703586950023,
     "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": 0.9046422840001469,
-      "bytes": 1080144,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.0989511400002812,
       "bytes": 7379943,
@@ -562,6 +556,12 @@
       "bytes": 847459,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.3560152030004247,
+      "bytes": 13557643,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -272,3 +272,4 @@
 {"step": 342200, "loss": 0.5539342202246189, "lr": 0.0003, "grad_norm": 0.1576419914149281, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50088846883, "tokens_per_second": 148006.31993308643, "data_wait_s": 1.4042001566849649e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 342300, "loss": 0.63454519957304, "lr": 0.0003, "grad_norm": 0.16010137369153826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50103505821, "tokens_per_second": 132307.1079772074, "data_wait_s": 1.2348999007372186e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 342400, "loss": 0.586083747446537, "lr": 0.0003, "grad_norm": 0.15601250614664455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50118267187, "tokens_per_second": 150083.05679639144, "data_wait_s": 1.2797998351743445e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 342500, "loss": 0.6427940670400858, "lr": 0.0003, "grad_norm": 0.15832013064999798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50132956989, "tokens_per_second": 149089.9116297351, "data_wait_s": 3.578399991965853e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0581c64c681710be5327baddb0cd11f91fdd98d1f8db63f45cca8aebecd0140
-size 584412
+oid sha256:3492ee47ee7a0655ef5d7ebc54c1c2312f45babac8142879074b91df219e0785
+size 586530