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": 76377.00855813155,
-      "required_pages_per_second_upper_estimate": 0.0052839386711274194
+      "required_token_positions_per_second": 76378.53485172753,
+      "required_pages_per_second_upper_estimate": 0.005284044263659878
     },
     {
       "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": 9547.126069766444,
-      "required_pages_per_second_upper_estimate": 0.003997182318748134
+      "required_token_positions_per_second": 9547.31685646594,
+      "required_pages_per_second_upper_estimate": 0.003997262197155135
     },
     {
       "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": 19094.252139532888,
-      "required_pages_per_second_upper_estimate": 0.003976829171373558
+      "required_token_positions_per_second": 19094.63371293188,
+      "required_pages_per_second_upper_estimate": 0.003976908643049801
     },
     {
       "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": 28641.37820929933,
-      "required_pages_per_second_upper_estimate": 0.0017170068907951836
+      "required_token_positions_per_second": 28641.950569397824,
+      "required_pages_per_second_upper_estimate": 0.001717041202909145
     },
     {
       "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": 28641.37820929933,
-      "required_pages_per_second_upper_estimate": 0.0022267411756642687
+      "required_token_positions_per_second": 28641.950569397824,
+      "required_pages_per_second_upper_estimate": 0.0022267856741443803
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28641.37820929933,
-      "required_pages_per_second_upper_estimate": 0.3673808469529551
+      "required_token_positions_per_second": 28641.950569397824,
+      "required_pages_per_second_upper_estimate": 0.36738818857374617
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10561898844532891,
-    "producer_p95_seconds": 0.21382130700112612,
-    "gpu_update_seconds": 0.9653166756836862,
-    "producer_batches_per_second": 9.46799448394287,
-    "gpu_batches_per_second": 1.0359294780562547,
-    "producer_headroom": 9.13961296063121,
+    "producer_batch_seconds": 0.07058400716400115,
+    "producer_p95_seconds": 0.18972024800132203,
+    "gpu_update_seconds": 0.9652973854909239,
+    "producer_batches_per_second": 14.167515279722094,
+    "gpu_batches_per_second": 1.035950179738058,
+    "producer_headroom": 13.675865458418452,
     "sustainable": true,
-    "buffer_seconds": 60.81495056807223,
+    "buffer_seconds": 60.81373528592821,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316940.90217190754,
-    "upload_bytes_per_second": 4787377.77284189,
+    "generation_bytes_per_second": 341621.94093631627,
+    "upload_bytes_per_second": 4870149.244763645,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.00813959949417,
+    "estimated_full_shards_per_hour": 73.30411597315899,
     "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": 1.4797563120000632,
-      "bytes": 958250,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.1572930810000344,
       "bytes": 7429293,
@@ -562,6 +556,12 @@
       "bytes": 11638802,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.3405057710006076,
+      "bytes": 7718368,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -80,3 +80,4 @@
 {"step": 850500, "loss": 0.4876767694950104, "lr": 3e-05, "grad_norm": 0.28623268586402095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124489747449, "tokens_per_second": 152333.05123555998, "data_wait_s": 9.347000741399825e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 850600, "loss": 0.5145455598831177, "lr": 3e-05, "grad_norm": 0.3018728292464876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124504502492, "tokens_per_second": 154343.01561946663, "data_wait_s": 1.0376999853178859e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 850700, "loss": 0.506303071975708, "lr": 3e-05, "grad_norm": 0.2909025878885348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124519048940, "tokens_per_second": 151697.43609105863, "data_wait_s": 9.502000466454774e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 850800, "loss": 0.6000729203224182, "lr": 3e-05, "grad_norm": 0.30744469547091996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124533588924, "tokens_per_second": 151344.06133344714, "data_wait_s": 8.895000064512715e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6529139bf970eff47f29e95b8be62d2898eb8b5ed180d20fe92382b345c908c
-size 175407
+oid sha256:9d48a2fef596a1698423d13d4e5c0db9a471c41ee0cd73773e0d89a7447bc8d5
+size 177525