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": 77131.86722534934,
-      "required_pages_per_second_upper_estimate": 0.005336161545238966
+      "required_token_positions_per_second": 77552.45949048121,
+      "required_pages_per_second_upper_estimate": 0.005365259094049298
     },
     {
       "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": 9641.483403168668,
-      "required_pages_per_second_upper_estimate": 0.004036687763838462
+      "required_token_positions_per_second": 9694.057436310151,
+      "required_pages_per_second_upper_estimate": 0.004058699413644146
     },
     {
       "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": 19282.966806337336,
-      "required_pages_per_second_upper_estimate": 0.0040161334597284895
+      "required_token_positions_per_second": 19388.114872620303,
+      "required_pages_per_second_upper_estimate": 0.004038033028994251
     },
     {
       "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": 28924.450209506,
-      "required_pages_per_second_upper_estimate": 0.0017339766249816558
+      "required_token_positions_per_second": 29082.172308930454,
+      "required_pages_per_second_upper_estimate": 0.0017434318240144504
     },
     {
       "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": 28924.450209506,
-      "required_pages_per_second_upper_estimate": 0.0022487487785781946
+      "required_token_positions_per_second": 29082.172308930454,
+      "required_pages_per_second_upper_estimate": 0.002261010978062248
     },
     {
       "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": 28924.450209506,
-      "required_pages_per_second_upper_estimate": 0.3710117906325727
+      "required_token_positions_per_second": 29082.172308930454,
+      "required_pages_per_second_upper_estimate": 0.37303488037519345
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.33146832416400684,
-    "producer_p95_seconds": 0.3058499509934336,
-    "gpu_update_seconds": 0.955869508313541,
-    "producer_batches_per_second": 3.0168795239246182,
-    "gpu_batches_per_second": 1.0461679039896556,
-    "producer_headroom": 2.8837431471750143,
+    "producer_batch_seconds": 0.3360371600391545,
+    "producer_p95_seconds": 0.3231841320011881,
+    "gpu_update_seconds": 0.9506855164154965,
+    "producer_batches_per_second": 2.9758613597480754,
+    "gpu_batches_per_second": 1.051872551682959,
+    "producer_headroom": 2.8291082935730207,
     "sustainable": true,
-    "buffer_seconds": 60.219779023753084,
+    "buffer_seconds": 60.843873050591775,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391123.2491378463,
-    "upload_bytes_per_second": 4897430.65850128,
+    "generation_bytes_per_second": 390964.5723708254,
+    "upload_bytes_per_second": 5119585.079307109,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.92594438172857,
+    "estimated_full_shards_per_hour": 83.89189604133198,
     "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": 3.7047144000025582,
-      "bytes": 6068293,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0213589359991602,
       "bytes": 2097340,
@@ -562,6 +556,12 @@
       "bytes": 901723,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.5148740549993818,
+      "bytes": 14069210,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -359,3 +359,4 @@
 {"step": 920900, "loss": 0.5707686543464661, "lr": 3e-05, "grad_norm": 0.35707305900982755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134793014607, "tokens_per_second": 153371.9813734807, "data_wait_s": 1.4352001016959548e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 921000, "loss": 0.5189375281333923, "lr": 3e-05, "grad_norm": 0.33550743073941924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134807539492, "tokens_per_second": 151413.46363908608, "data_wait_s": 1.3078999472782016e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 921100, "loss": 0.6511127948760986, "lr": 3e-05, "grad_norm": 0.3860076778193839, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134822359174, "tokens_per_second": 154537.08573383294, "data_wait_s": 1.3713994121644646e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 921200, "loss": 0.5252535939216614, "lr": 3e-05, "grad_norm": 0.3546703675861652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134837222360, "tokens_per_second": 153886.84473517977, "data_wait_s": 1.3698998373001814e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8a7f99d23bd97d66b31240fe9539bee35a3ddcaab9ef1bf882e51b3ab7f263d
-size 770244
+oid sha256:90598712fc3bd41283f5c0852344f7707133a86fbb841e06879c1899aa17e339
+size 772362