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": 76034.98163354839,
-      "required_pages_per_second_upper_estimate": 0.0065905194990968545
+      "required_token_positions_per_second": 75901.69469994892,
+      "required_pages_per_second_upper_estimate": 0.006578966525505097
     },
     {
       "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": 9504.372704193549,
-      "required_pages_per_second_upper_estimate": 0.003791436374738132
+      "required_token_positions_per_second": 9487.711837493614,
+      "required_pages_per_second_upper_estimate": 0.003784790105909372
     },
     {
       "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": 19008.745408387098,
-      "required_pages_per_second_upper_estimate": 0.004397338707737212
+      "required_token_positions_per_second": 18975.42367498723,
+      "required_pages_per_second_upper_estimate": 0.004389630311157631
     },
     {
       "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": 28513.118112580647,
-      "required_pages_per_second_upper_estimate": 0.0017362992962868722
+      "required_token_positions_per_second": 28463.135512480843,
+      "required_pages_per_second_upper_estimate": 0.00173325561817923
     },
     {
       "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": 28513.118112580647,
-      "required_pages_per_second_upper_estimate": 0.002195018761669098
+      "required_token_positions_per_second": 28463.135512480843,
+      "required_pages_per_second_upper_estimate": 0.0021911709627527223
     },
     {
       "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": 28513.118112580647,
-      "required_pages_per_second_upper_estimate": 0.354037499690585
+      "required_token_positions_per_second": 28463.135512480843,
+      "required_pages_per_second_upper_estimate": 0.3534168830783471
     }
   ],
   "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": 0.055721478953124404,
-    "producer_p95_seconds": 0.198875637999663,
-    "gpu_update_seconds": 0.96965894402833,
-    "producer_batches_per_second": 17.946400899395513,
-    "gpu_batches_per_second": 1.0312904409932238,
-    "producer_headroom": 17.401888145216926,
+    "producer_batch_seconds": 0.0708772579062753,
+    "producer_p95_seconds": 0.20240832699983002,
+    "gpu_update_seconds": 0.971361710584436,
+    "producer_batches_per_second": 14.108897967276784,
+    "gpu_batches_per_second": 1.0294826212558175,
+    "producer_headroom": 13.704843263955249,
     "sustainable": true,
-    "buffer_seconds": 60.11885452975646,
+    "buffer_seconds": 60.22442605623503,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357958.93010807445,
-    "upload_bytes_per_second": 3288607.199507182,
+    "generation_bytes_per_second": 342959.2492229392,
+    "upload_bytes_per_second": 3374502.300141641,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.80965354377437,
+    "estimated_full_shards_per_hour": 73.59107120052464,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 1122627,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.176693084999897,
+      "bytes": 12183576,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -59,3 +59,4 @@
 {"step": 243000, "loss": 0.5930889509618282, "lr": 0.0003, "grad_norm": 0.1354303855648867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35567657221, "tokens_per_second": 150613.2766396233, "data_wait_s": 1.6164000044227578e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 243100, "loss": 0.5773447453975677, "lr": 0.0003, "grad_norm": 0.14863892524941433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35582463245, "tokens_per_second": 151016.43599002584, "data_wait_s": 1.1258000085945241e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 243200, "loss": 0.6425658594816923, "lr": 0.0003, "grad_norm": 0.15166356932682554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35597037777, "tokens_per_second": 148179.4925929846, "data_wait_s": 2.8018999728374183e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 243300, "loss": 0.591025285422802, "lr": 0.0003, "grad_norm": 0.1752452655497461, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35611473614, "tokens_per_second": 147379.8531822967, "data_wait_s": 1.3977999515191186e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ab301d52e2ca3ce7c01fb497477689278ed71cb98b984507baabf365b625d6f
-size 130146
+oid sha256:4d9ef0a10f8dd153e770940bc7f1a5ce15367542143adbe3046bbd6b1a9221ea
+size 132264