Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76225.64318975309,
-      "required_pages_per_second_upper_estimate": 0.005273466863726955
+      "required_token_positions_per_second": 75672.04416088146,
+      "required_pages_per_second_upper_estimate": 0.005235167598382895
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9528.205398719136,
-      "required_pages_per_second_upper_estimate": 0.003989260628889167
+      "required_token_positions_per_second": 9459.005520110182,
+      "required_pages_per_second_upper_estimate": 0.003960288084773386
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19056.41079743827,
-      "required_pages_per_second_upper_estimate": 0.003968947817758549
+      "required_token_positions_per_second": 18918.011040220365,
+      "required_pages_per_second_upper_estimate": 0.003940122798177098
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28584.616196157407,
-      "required_pages_per_second_upper_estimate": 0.0017136040947779009
+      "required_token_positions_per_second": 28377.016560330543,
+      "required_pages_per_second_upper_estimate": 0.0017011588135963703
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28584.616196157407,
-      "required_pages_per_second_upper_estimate": 0.0022223281788121938
+      "required_token_positions_per_second": 28377.016560330543,
+      "required_pages_per_second_upper_estimate": 0.00220618822026095
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28584.616196157407,
-      "required_pages_per_second_upper_estimate": 0.3666527647946718
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28377.016560330543,
+      "required_pages_per_second_upper_estimate": 0.363989899569407
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.33373548661705854,
-    "producer_p95_seconds": 0.3024194300014642,
-    "gpu_update_seconds": 0.9672335570388623,
-    "producer_batches_per_second": 2.9963849818207677,
-    "gpu_batches_per_second": 1.0338764538540728,
-    "producer_headroom": 2.8982041042243276,
+    "producer_batch_seconds": 0.10274186181266032,
+    "producer_p95_seconds": 0.2883333780009707,
+    "gpu_update_seconds": 0.9743096121898287,
+    "producer_batches_per_second": 9.73313099798991,
+    "gpu_batches_per_second": 1.0263677864702887,
+    "producer_headroom": 9.48308308804435,
     "sustainable": true,
-    "buffer_seconds": 60.935714093448325,
+    "buffer_seconds": 60.40719595576938,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316701.839399273,
-    "upload_bytes_per_second": 5044749.677723346,
+    "generation_bytes_per_second": 328559.36564273876,
+    "upload_bytes_per_second": 5159175.69455878,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.95684229358332,
+    "estimated_full_shards_per_hour": 70.50119139634725,
     "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": 3.168707209999411,
-      "bytes": 6305368,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.9134112159990764,
       "bytes": 11699377,
@@ -562,6 +556,12 @@
       "bytes": 7432740,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1428929149988107,
+      "bytes": 5656732,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 951200, "loss": 0.5129421353340149, "lr": 3e-05, "grad_norm": 0.33315655108107156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139229666414, "tokens_per_second": 150673.39371447515, "data_wait_s": 1.5237001207424328e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 951300, "loss": 0.5131524205207825, "lr": 3e-05, "grad_norm": 0.34254456535394895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139244193554, "tokens_per_second": 151113.32410119823, "data_wait_s": 1.3173998013371602e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 951400, "loss": 0.5310010313987732, "lr": 3e-05, "grad_norm": 0.3356313779004489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139258878320, "tokens_per_second": 152646.37032217704, "data_wait_s": 1.300000076298602e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 951500, "loss": 0.5428860187530518, "lr": 3e-05, "grad_norm": 0.3392346283253021, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139273610388, "tokens_per_second": 153114.31166177063, "data_wait_s": 1.4474000636255369e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:28c811a9fa9b183dc65869eba23b739f8f1e62b6141d9f81bfba9fb590aa45c4
-size 563232
+oid sha256:03b1c30d4bd205f60e84e77271ea9cd693423e7ad8cb1d4c9881003c25daaafc
+size 565350