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": 74101.0507728367,
-      "required_pages_per_second_upper_estimate": 0.00642289127359332
+      "required_token_positions_per_second": 74857.21219245231,
+      "required_pages_per_second_upper_estimate": 0.006488433428972005
     },
     {
       "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": 9262.631346604587,
-      "required_pages_per_second_upper_estimate": 0.0036950021328405083
+      "required_token_positions_per_second": 9357.151524056539,
+      "required_pages_per_second_upper_estimate": 0.0037327076448286814
     },
     {
       "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": 18525.262693209173,
-      "required_pages_per_second_upper_estimate": 0.004285493490585968
+      "required_token_positions_per_second": 18714.303048113077,
+      "required_pages_per_second_upper_estimate": 0.004329224649696371
     },
     {
       "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": 27787.89403981376,
-      "required_pages_per_second_upper_estimate": 0.0016921369552120162
+      "required_token_positions_per_second": 28071.454572169616,
+      "required_pages_per_second_upper_estimate": 0.001709404303905352
     },
     {
       "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": 27787.89403981376,
-      "required_pages_per_second_upper_estimate": 0.0021391890049987806
+      "required_token_positions_per_second": 28071.454572169616,
+      "required_pages_per_second_upper_estimate": 0.002161018279725327
     },
     {
       "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": 27787.89403981376,
-      "required_pages_per_second_upper_estimate": 0.34503264387565663
+      "required_token_positions_per_second": 28071.454572169616,
+      "required_pages_per_second_upper_estimate": 0.348553516671463
     }
   ],
   "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": 0.3641113040937398,
-    "producer_p95_seconds": 0.21264524300022458,
-    "gpu_update_seconds": 0.9949656480043675,
-    "producer_batches_per_second": 2.7464129477906893,
-    "gpu_batches_per_second": 1.0050598249353935,
-    "producer_headroom": 2.732586538286148,
+    "producer_batch_seconds": 0.05575986999220106,
+    "producer_p95_seconds": 0.18823832199996104,
+    "gpu_update_seconds": 0.9849151182714474,
+    "producer_batches_per_second": 17.934044683745974,
+    "gpu_batches_per_second": 1.0153159205790514,
+    "producer_headroom": 17.663511740777086,
     "sustainable": true,
-    "buffer_seconds": 60.69290452826642,
+    "buffer_seconds": 60.07982221455829,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300070.1614429042,
+    "generation_bytes_per_second": 299012.9037481205,
     "upload_bytes_per_second": 6039631.556071958,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.38807136979432,
+    "estimated_full_shards_per_hour": 64.16120848019324,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -65,3 +65,4 @@
 {"step": 91500, "loss": 0.6084352731704712, "lr": 0.0003, "grad_norm": 0.12859608848031054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13390574656, "tokens_per_second": 148032.37163812228, "data_wait_s": 1.0539999493630603e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 91600, "loss": 0.6036572568118572, "lr": 0.0003, "grad_norm": 0.1329230079673119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13405010480, "tokens_per_second": 146787.359406769, "data_wait_s": 1.424999936716631e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 91700, "loss": 0.6243661977350712, "lr": 0.0003, "grad_norm": 0.1397994986921189, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13419543507, "tokens_per_second": 148050.62269793818, "data_wait_s": 1.1871999959112145e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 91800, "loss": 0.5888327732682228, "lr": 0.0003, "grad_norm": 0.12957206124342208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13434264611, "tokens_per_second": 150189.06462793087, "data_wait_s": 1.0155999916605651e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:379ee64ab887144622ea6ec94e37a65ba710df5910b770a3badb1cf7dec9fad9
-size 142854
+oid sha256:ff5cd018d530058868e6ea0ba2446069b4bfb4a31d74a38c7dd2f6783bf8b3a0
+size 144972