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": 76014.41205826009,
-      "required_pages_per_second_upper_estimate": 0.006588736580443976
+      "required_token_positions_per_second": 75682.30749656633,
+      "required_pages_per_second_upper_estimate": 0.0065599505987476775
     },
     {
       "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": 9501.801507282511,
-      "required_pages_per_second_upper_estimate": 0.003790410685847499
+      "required_token_positions_per_second": 9460.28843707079,
+      "required_pages_per_second_upper_estimate": 0.003773850501464333
     },
     {
       "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": 19003.603014565022,
-      "required_pages_per_second_upper_estimate": 0.004396149105429515
+      "required_token_positions_per_second": 18920.57687414158,
+      "required_pages_per_second_upper_estimate": 0.004376942469052718
     },
     {
       "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": 28505.40452184753,
-      "required_pages_per_second_upper_estimate": 0.0017358295790813082
+      "required_token_positions_per_second": 28380.865311212372,
+      "required_pages_per_second_upper_estimate": 0.0017282457945603674
     },
     {
       "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": 28505.40452184753,
-      "required_pages_per_second_upper_estimate": 0.0021944249481018765
+      "required_token_positions_per_second": 28380.865311212372,
+      "required_pages_per_second_upper_estimate": 0.0021848375749205823
     },
     {
       "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": 28505.40452184753,
-      "required_pages_per_second_upper_estimate": 0.3539417227094072
+      "required_token_positions_per_second": 28380.865311212372,
+      "required_pages_per_second_upper_estimate": 0.35239536251924425
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.10243860142958283,
-    "producer_p95_seconds": 0.3000443639975856,
-    "gpu_update_seconds": 0.9699213347002185,
-    "producer_batches_per_second": 9.761945068016265,
-    "gpu_batches_per_second": 1.0310114482728419,
-    "producer_headroom": 9.46831878964055,
+    "producer_batch_seconds": 0.08432810068734398,
+    "producer_p95_seconds": 0.2457931009994354,
+    "gpu_update_seconds": 0.9741774853170936,
+    "producer_batches_per_second": 11.858443292913874,
+    "gpu_batches_per_second": 1.0265069918696603,
+    "producer_headroom": 11.552228466866191,
     "sustainable": true,
-    "buffer_seconds": 60.13512275141355,
+    "buffer_seconds": 60.399004089659805,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354913.64208096,
-    "upload_bytes_per_second": 5696011.509067715,
+    "generation_bytes_per_second": 327079.61495082325,
+    "upload_bytes_per_second": 5751618.036218131,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.1562056238327,
+    "estimated_full_shards_per_hour": 70.18367134469531,
     "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": 1.0303949059998558,
-      "bytes": 5043636,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.2701704230003088,
       "bytes": 6145157,
@@ -562,6 +556,12 @@
       "bytes": 2615219,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2512058009997418,
+      "bytes": 10378380,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -214,3 +214,4 @@
 {"step": 376400, "loss": 0.594700651243329, "lr": 0.0003, "grad_norm": 0.14515521444552787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55093777204, "tokens_per_second": 149790.23010791634, "data_wait_s": 1.3443001080304384e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 376500, "loss": 0.6773975044488907, "lr": 0.0003, "grad_norm": 0.20503113867207742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55108505761, "tokens_per_second": 150725.35792295678, "data_wait_s": 1.1152002116432413e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 376600, "loss": 0.5772315664216876, "lr": 0.0003, "grad_norm": 0.150786749470871, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55123198148, "tokens_per_second": 142343.11339296712, "data_wait_s": 1.4399000065168366e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 376700, "loss": 0.583888977766037, "lr": 0.0003, "grad_norm": 0.16854099997862482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55137931768, "tokens_per_second": 150553.43137612595, "data_wait_s": 2.1908999769948423e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:081732399f429842656730a94e62d7ba71023b4bbfcdbc28f1eeed145f04a390
-size 460785
+oid sha256:0b8ace0e5da92985ff2f0208ed7ce773a0553596a79b7fee79df6f853167f432
+size 462903