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": 75401.04768176952,
-      "required_pages_per_second_upper_estimate": 0.006535571710847574
+      "required_token_positions_per_second": 75643.02484417692,
+      "required_pages_per_second_upper_estimate": 0.006556545677999536
     },
     {
       "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": 9425.13096022119,
-      "required_pages_per_second_upper_estimate": 0.0037598256582979054
+      "required_token_positions_per_second": 9455.378105522115,
+      "required_pages_per_second_upper_estimate": 0.0037718916967935676
     },
     {
       "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": 18850.26192044238,
-      "required_pages_per_second_upper_estimate": 0.004360676342015326
+      "required_token_positions_per_second": 18910.75621104423,
+      "required_pages_per_second_upper_estimate": 0.004374670631482913
     },
     {
       "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": 28275.392880663567,
-      "required_pages_per_second_upper_estimate": 0.0017218230769110215
+      "required_token_positions_per_second": 28366.134316566342,
+      "required_pages_per_second_upper_estimate": 0.00172734875427398
     },
     {
       "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": 28275.392880663567,
-      "required_pages_per_second_upper_estimate": 0.002176718015250561
+      "required_token_positions_per_second": 28366.134316566342,
+      "required_pages_per_second_upper_estimate": 0.002183703542174728
     },
     {
       "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": 28275.392880663567,
-      "required_pages_per_second_upper_estimate": 0.3510857479253457
+      "required_token_positions_per_second": 28366.134316566342,
+      "required_pages_per_second_upper_estimate": 0.35221245286720815
     }
   ],
   "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": 0.06868094467191099,
-    "producer_p95_seconds": 0.2551642409998749,
-    "gpu_update_seconds": 0.9778113470142933,
-    "producer_batches_per_second": 14.560079288032538,
-    "gpu_batches_per_second": 1.0226921614823339,
-    "producer_headroom": 14.237010741266008,
+    "producer_batch_seconds": 0.08517846446869726,
+    "producer_p95_seconds": 0.22615953400236322,
+    "gpu_update_seconds": 0.9746833915206085,
+    "producer_batches_per_second": 11.74005667086774,
+    "gpu_batches_per_second": 1.0259741867971044,
+    "producer_headroom": 11.442838252605512,
     "sustainable": true,
-    "buffer_seconds": 60.624303514886186,
+    "buffer_seconds": 60.43037027427773,
     "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": 323855.48079424293,
-    "upload_bytes_per_second": 5838699.178988453,
+    "generation_bytes_per_second": 326584.9894972917,
+    "upload_bytes_per_second": 5785835.687402418,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.49184720869509,
+    "estimated_full_shards_per_hour": 70.0775362366587,
     "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": 1.2508507479997206,
-      "bytes": 11653692,
-      "pages": 45,
-      "shards": 4
-    },
     {
       "seconds": 1.2195578610007942,
       "bytes": 5944204,
@@ -562,6 +556,12 @@
       "bytes": 7534736,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.0304640750000544,
+      "bytes": 6585841,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -176,3 +176,4 @@
 {"step": 332600, "loss": 0.582495553418994, "lr": 0.0003, "grad_norm": 0.14115508804723378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48684481372, "tokens_per_second": 129070.67810782626, "data_wait_s": 1.3593999028671533e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 332700, "loss": 0.6097974088042974, "lr": 0.0003, "grad_norm": 0.1485296390404382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48699049698, "tokens_per_second": 148018.8358762755, "data_wait_s": 1.3437998859444633e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 332800, "loss": 0.6080064345151186, "lr": 0.0003, "grad_norm": 0.1539737111117055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48713588007, "tokens_per_second": 148049.72459891406, "data_wait_s": 1.3767999917035922e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 332900, "loss": 0.6819286067038774, "lr": 0.0003, "grad_norm": 0.2294065775638761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48728128703, "tokens_per_second": 146969.44527870012, "data_wait_s": 1.3622000551549718e-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:bbeb375802719f15edd12146450a6eaf67c6c6d935ae625a0726fbd06cb98573
-size 379518
+oid sha256:7bbc62bb6097b3417e183b7994ba9900525004937f7e557ce1e04332530aea9f
+size 381636