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": 75362.42562393697,
-      "required_pages_per_second_upper_estimate": 0.006532224048761359
+      "required_token_positions_per_second": 76039.83842698619,
+      "required_pages_per_second_upper_estimate": 0.006590940473642612
     },
     {
       "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": 9420.303202992121,
-      "required_pages_per_second_upper_estimate": 0.003757899793757827
+      "required_token_positions_per_second": 9504.979803373273,
+      "required_pages_per_second_upper_estimate": 0.0037916785556778654
     },
     {
       "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": 18840.606405984243,
-      "required_pages_per_second_upper_estimate": 0.0043584427086766865
+      "required_token_positions_per_second": 19009.959606746546,
+      "required_pages_per_second_upper_estimate": 0.0043976195911585095
     },
     {
       "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": 28260.909608976366,
-      "required_pages_per_second_upper_estimate": 0.0017209411216531248
+      "required_token_positions_per_second": 28514.939410119823,
+      "required_pages_per_second_upper_estimate": 0.0017364102037513984
     },
     {
       "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": 28260.909608976366,
-      "required_pages_per_second_upper_estimate": 0.0021756030529038184
+      "required_token_positions_per_second": 28514.939410119823,
+      "required_pages_per_second_upper_estimate": 0.0021951589701953377
     },
     {
       "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": 28260.909608976366,
-      "required_pages_per_second_upper_estimate": 0.35090591416338285
+      "required_token_positions_per_second": 28514.939410119823,
+      "required_pages_per_second_upper_estimate": 0.35406011411053084
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.4923171177578638,
-    "producer_p95_seconds": 0.2890667060000851,
-    "gpu_update_seconds": 0.9783124599506277,
-    "producer_batches_per_second": 2.031211111558038,
-    "gpu_batches_per_second": 1.0221683162968882,
-    "producer_headroom": 1.9871591392273928,
+    "producer_batch_seconds": 0.09390910780474826,
+    "producer_p95_seconds": 0.2217405269984738,
+    "gpu_update_seconds": 0.9695970102671111,
+    "producer_batches_per_second": 10.648594405551766,
+    "gpu_batches_per_second": 1.0313563154701904,
+    "producer_headroom": 10.324845299170077,
     "sustainable": true,
-    "buffer_seconds": 60.65537251693891,
+    "buffer_seconds": 60.115014636560886,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309521.43299702473,
-    "upload_bytes_per_second": 5874854.340514966,
+    "generation_bytes_per_second": 317395.7115006348,
+    "upload_bytes_per_second": 5895625.720581957,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.4160942309671,
+    "estimated_full_shards_per_hour": 68.10573109401973,
     "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": 1.2743502019993684,
-      "bytes": 11373993,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 0.9631920529991476,
       "bytes": 874408,
@@ -562,6 +556,12 @@
       "bytes": 652793,
       "pages": 19,
       "shards": 1
+    },
+    {
+      "seconds": 1.3964870420004445,
+      "bytes": 13572779,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -140,3 +140,4 @@
 {"step": 129000, "loss": 0.5921800285577774, "lr": 0.0003, "grad_norm": 0.14160159800155678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18878604005, "tokens_per_second": 149773.38357872202, "data_wait_s": 1.458100086892955e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 129100, "loss": 0.5809275228530169, "lr": 0.0003, "grad_norm": 0.1398279316633142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18893266134, "tokens_per_second": 135279.68346460993, "data_wait_s": 2.1052999727544375e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 129200, "loss": 0.6225546598434448, "lr": 0.0003, "grad_norm": 0.1455784107006561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18907853512, "tokens_per_second": 149300.0929345801, "data_wait_s": 9.02399915503338e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 129300, "loss": 0.6238899454474449, "lr": 0.0003, "grad_norm": 0.15132003403086974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18922486391, "tokens_per_second": 149880.6559437037, "data_wait_s": 1.3660999684361741e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1144e9803735cf8466019db4f952f9f478c6cbacf6d4b69b386f224ae5b686f5
-size 302487
+oid sha256:a13b81b8941e19f82df0358a9c860118e3339bfec1bb53f372530b9f7ecf3045
+size 304605