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": 75668.30560624716,
-      "required_pages_per_second_upper_estimate": 0.006558736950382279
+      "required_token_positions_per_second": 75785.75275182704,
+      "required_pages_per_second_upper_estimate": 0.006568916971294073
     },
     {
       "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": 9458.538200780895,
-      "required_pages_per_second_upper_estimate": 0.003773152306039929
+      "required_token_positions_per_second": 9473.21909397838,
+      "required_pages_per_second_upper_estimate": 0.0037790087338353197
     },
     {
       "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": 18917.07640156179,
-      "required_pages_per_second_upper_estimate": 0.004376132696327596
+      "required_token_positions_per_second": 18946.43818795676,
+      "required_pages_per_second_upper_estimate": 0.004382925028860281
     },
     {
       "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": 28375.614602342685,
-      "required_pages_per_second_upper_estimate": 0.0017279260539385438
+      "required_token_positions_per_second": 28419.65728193514,
+      "required_pages_per_second_upper_estimate": 0.0017306080220516429
     },
     {
       "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": 28375.614602342685,
-      "required_pages_per_second_upper_estimate": 0.0021844333608169015
+      "required_token_positions_per_second": 28419.65728193514,
+      "required_pages_per_second_upper_estimate": 0.0021878238882098687
     },
     {
       "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": 28375.614602342685,
-      "required_pages_per_second_upper_estimate": 0.35233016628807484
+      "required_token_positions_per_second": 28419.65728193514,
+      "required_pages_per_second_upper_estimate": 0.3528770289672466
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.13612086404683055,
-    "producer_p95_seconds": 0.3552577820009901,
-    "gpu_update_seconds": 0.9743577500420867,
-    "producer_batches_per_second": 7.346412374049899,
-    "gpu_batches_per_second": 1.026317079077788,
-    "producer_headroom": 7.158033831660605,
+    "producer_batch_seconds": 0.09178190506256101,
+    "producer_p95_seconds": 0.36843975099691306,
+    "gpu_update_seconds": 0.972847762579261,
+    "producer_batches_per_second": 10.895393806855209,
+    "gpu_batches_per_second": 1.0279100579403624,
+    "producer_headroom": 10.599559487419027,
     "sustainable": true,
-    "buffer_seconds": 60.41018050260937,
+    "buffer_seconds": 60.31656127991418,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321608.0193672334,
+    "generation_bytes_per_second": 321661.0006089142,
     "upload_bytes_per_second": 4200074.488290216,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.00959430468322,
+    "estimated_full_shards_per_hour": 69.02096284580773,
     "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": [

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

@@ -299,3 +299,4 @@
 {"step": 344900, "loss": 0.5837528426200151, "lr": 0.0003, "grad_norm": 0.18961747227709821, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50483965852, "tokens_per_second": 148048.0188297243, "data_wait_s": 1.8357000953983516e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 345000, "loss": 0.5730580352246761, "lr": 0.0003, "grad_norm": 0.16349241880176352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50498610481, "tokens_per_second": 149348.46023168947, "data_wait_s": 1.3778000720776618e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 345100, "loss": 0.5767875835299492, "lr": 0.0003, "grad_norm": 0.14671507177890084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50513293229, "tokens_per_second": 133742.00070424526, "data_wait_s": 1.2344000424491242e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 345200, "loss": 0.5913329981267452, "lr": 0.0003, "grad_norm": 0.14752144068889433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50527916156, "tokens_per_second": 148072.05477990818, "data_wait_s": 1.4799999917158857e-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:fbcd2c57ee6ca95bd5146380f37e92cc759a574a92d77c27b1935009b7ead04e
-size 642381
+oid sha256:be2a13238789c8ebd1fe040ea7b0c9227eba39b6fb1dbf71ca1a821e040119cc
+size 644499