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": 74886.39024463865,
-      "required_pages_per_second_upper_estimate": 0.006490962508584419
+      "required_token_positions_per_second": 74552.94501767203,
+      "required_pages_per_second_upper_estimate": 0.006462060321420154
     },
     {
       "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": 9360.798780579831,
-      "required_pages_per_second_upper_estimate": 0.0037341625899871673
+      "required_token_positions_per_second": 9319.118127209003,
+      "required_pages_per_second_upper_estimate": 0.003717535554176242
     },
     {
       "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": 18721.597561159662,
-      "required_pages_per_second_upper_estimate": 0.004330912106910653
+      "required_token_positions_per_second": 18638.236254418007,
+      "required_pages_per_second_upper_estimate": 0.004311627935704862
     },
     {
       "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": 28082.39634173949,
-      "required_pages_per_second_upper_estimate": 0.0017100706002651292
+      "required_token_positions_per_second": 27957.35438162701,
+      "required_pages_per_second_upper_estimate": 0.001702456200938742
     },
     {
       "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": 28082.39634173949,
-      "required_pages_per_second_upper_estimate": 0.0021618606074355726
+      "required_token_positions_per_second": 27957.35438162701,
+      "required_pages_per_second_upper_estimate": 0.002152234531207814
     },
     {
       "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": 28082.39634173949,
-      "required_pages_per_second_upper_estimate": 0.3486893768297738
+      "required_token_positions_per_second": 27957.35438162701,
+      "required_pages_per_second_upper_estimate": 0.3471367741736967
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.18412051574993882,
-    "producer_p95_seconds": 0.2220428999989963,
-    "gpu_update_seconds": 0.9845313649001585,
-    "producer_batches_per_second": 5.4312252815875155,
-    "gpu_batches_per_second": 1.0157116732399991,
-    "producer_headroom": 5.347211639561604,
+    "producer_batch_seconds": 0.1109533526328903,
+    "producer_p95_seconds": 0.247826619000989,
+    "gpu_update_seconds": 0.9889347762522798,
+    "producer_batches_per_second": 9.01279660569325,
+    "gpu_batches_per_second": 1.0111890329002824,
+    "producer_headroom": 8.913067994658562,
     "sustainable": true,
-    "buffer_seconds": 60.056413258909664,
+    "buffer_seconds": 60.32502135138907,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 447512.6218550884,
-    "upload_bytes_per_second": 5285081.136332803,
+    "generation_bytes_per_second": 444322.4518876011,
+    "upload_bytes_per_second": 5202742.941964504,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.02579108943452,
+    "estimated_full_shards_per_hour": 95.3412548777678,
     "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": 0.9735736169996017,
-      "bytes": 6894198,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.282220471999608,
       "bytes": 9425496,
@@ -562,6 +556,12 @@
       "bytes": 10461677,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.3869329340013792,
+      "bytes": 2462257,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -115,3 +115,4 @@
 {"step": 206500, "loss": 0.5908651724457741, "lr": 0.0003, "grad_norm": 0.1655434847203432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30223869542, "tokens_per_second": 147681.66956276508, "data_wait_s": 1.7620999642531388e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 206600, "loss": 0.5545372609049082, "lr": 0.0003, "grad_norm": 0.1425858499947372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30238319105, "tokens_per_second": 145724.7558609718, "data_wait_s": 1.7537000530865043e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 206700, "loss": 0.5890518985688686, "lr": 0.0003, "grad_norm": 0.1374137945616634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30252925621, "tokens_per_second": 146864.2398549859, "data_wait_s": 2.1116999050718732e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 206800, "loss": 0.5777712427079678, "lr": 0.0003, "grad_norm": 0.13962449979279243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30267646902, "tokens_per_second": 147041.5544525577, "data_wait_s": 1.715100006549619e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67c36c5f5861e6b94d716d56bef8fa279e1089ff51c241f88dcf4e3155e2f3f3
-size 249537
+oid sha256:324e3a2a4356cf9779f1d2b2289a51f3b91c34f74e117596e4dccc0a6785c4e2
+size 251655