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": 74783.973714564,
-      "required_pages_per_second_upper_estimate": 0.00648208530332987
+      "required_token_positions_per_second": 75228.61760565774,
+      "required_pages_per_second_upper_estimate": 0.006520625908870237
     },
     {
       "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": 9347.9967143205,
-      "required_pages_per_second_upper_estimate": 0.003729055654348372
+      "required_token_positions_per_second": 9403.577200707217,
+      "required_pages_per_second_upper_estimate": 0.0037512275413703598
     },
     {
       "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": 18695.993428641,
-      "required_pages_per_second_upper_estimate": 0.004324989041469803
+      "required_token_positions_per_second": 18807.154401414435,
+      "required_pages_per_second_upper_estimate": 0.004350704176154635
     },
     {
       "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": 28043.9901429615,
-      "required_pages_per_second_upper_estimate": 0.0017077318642613015
+      "required_token_positions_per_second": 28210.73160212165,
+      "required_pages_per_second_upper_estimate": 0.0017178855442993286
     },
     {
       "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": 28043.9901429615,
-      "required_pages_per_second_upper_estimate": 0.00215890399193848
+      "required_token_positions_per_second": 28210.73160212165,
+      "required_pages_per_second_upper_estimate": 0.002171740211034529
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28043.9901429615,
-      "required_pages_per_second_upper_estimate": 0.34821250037822987
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28210.73160212165,
+      "required_pages_per_second_upper_estimate": 0.3502828712532333
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.059340701453081124,
-    "producer_p95_seconds": 0.20147364500007825,
-    "gpu_update_seconds": 0.9858796789992139,
-    "producer_batches_per_second": 16.851839892567995,
-    "gpu_batches_per_second": 1.0143225601476236,
-    "producer_headroom": 16.613886503831083,
+    "producer_batch_seconds": 0.11253154135938814,
+    "producer_p95_seconds": 0.21601555200140865,
+    "gpu_update_seconds": 0.9800525697079289,
+    "producer_batches_per_second": 8.886397430621999,
+    "gpu_batches_per_second": 1.0203534288961824,
+    "producer_headroom": 8.709136637327026,
     "sustainable": true,
-    "buffer_seconds": 60.13866041895205,
+    "buffer_seconds": 60.76325932189159,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315655.06485452,
-    "upload_bytes_per_second": 5234108.162709917,
+    "generation_bytes_per_second": 353505.4464497467,
+    "upload_bytes_per_second": 5276205.103530173,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.7322288439436,
+    "estimated_full_shards_per_hour": 75.8540396224909,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 5623378,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.9634915199985699,
+      "bytes": 7597611,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -86,3 +86,4 @@
 {"step": 203600, "loss": 0.5829173810780048, "lr": 0.0003, "grad_norm": 0.15844119669134157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29799460784, "tokens_per_second": 148556.21699054367, "data_wait_s": 1.7874001059681177e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203700, "loss": 0.6329334229230881, "lr": 0.0003, "grad_norm": 0.1702569485568475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29814096252, "tokens_per_second": 147765.11220061584, "data_wait_s": 1.8253998860018328e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203800, "loss": 0.5466072801500559, "lr": 0.0003, "grad_norm": 0.14604799335517452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29828659819, "tokens_per_second": 146933.57037424322, "data_wait_s": 2.0323001081123948e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 203900, "loss": 0.5432954356074333, "lr": 0.0003, "grad_norm": 0.15455388367935946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29843379167, "tokens_per_second": 147760.30112497802, "data_wait_s": 1.8421000277157873e-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:ef1d2413b2f396b9fbd69097ba72e485950ef642a8052c5990dcd7a278044368
-size 187332
+oid sha256:a020733d9a8808d2dbf0ac47da0750b59c81705a13ec91d06f402b90b62ecad5
+size 189450