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": 77148.07643973226,
-      "required_pages_per_second_upper_estimate": 0.005337282936300535
+      "required_token_positions_per_second": 75018.40766475233,
+      "required_pages_per_second_upper_estimate": 0.0051899475089350585
     },
     {
       "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": 9643.509554966533,
-      "required_pages_per_second_upper_estimate": 0.004037536071285367
+      "required_token_positions_per_second": 9377.300958094042,
+      "required_pages_per_second_upper_estimate": 0.0039260800908425
     },
     {
       "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": 19287.019109933066,
-      "required_pages_per_second_upper_estimate": 0.004016977447701048
+      "required_token_positions_per_second": 18754.601916188083,
+      "required_pages_per_second_upper_estimate": 0.003906088987029569
     },
     {
       "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": 28930.5286648996,
-      "required_pages_per_second_upper_estimate": 0.001734341019101238
+      "required_token_positions_per_second": 28131.90287428213,
+      "required_pages_per_second_upper_estimate": 0.0016864646223846956
     },
     {
       "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": 28930.5286648996,
-      "required_pages_per_second_upper_estimate": 0.0022492213517487473
+      "required_token_positions_per_second": 28131.90287428213,
+      "required_pages_per_second_upper_estimate": 0.0021871317093118494
     },
     {
       "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": 28930.5286648996,
-      "required_pages_per_second_upper_estimate": 0.3710897585318249
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28131.90287428213,
+      "required_pages_per_second_upper_estimate": 0.3608458443873492
     }
   ],
   "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": 0.06000040915633065,
-    "producer_p95_seconds": 0.20011380000141799,
-    "gpu_update_seconds": 0.9556686751301697,
-    "producer_batches_per_second": 16.66655301290541,
-    "gpu_batches_per_second": 1.0463877555302228,
-    "producer_headroom": 15.927702636830054,
+    "producer_batch_seconds": 0.596125458539106,
+    "producer_p95_seconds": 0.2042617999995855,
+    "gpu_update_seconds": 0.9827987862589805,
+    "producer_batches_per_second": 1.6774992338871897,
+    "gpu_batches_per_second": 1.0175022740987458,
+    "producer_headroom": 1.6486442110146997,
     "sustainable": true,
-    "buffer_seconds": 60.20712653320069,
+    "buffer_seconds": 60.93352474805679,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 282002.3070192679,
-    "upload_bytes_per_second": 4815102.240921509,
+    "generation_bytes_per_second": 239188.33779323788,
+    "upload_bytes_per_second": 4805205.1694681635,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.51113040860679,
+    "estimated_full_shards_per_hour": 51.3242492708955,
     "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": 1.523682868999913,
-      "bytes": 6058267,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.493248520999714,
       "bytes": 2669063,
@@ -562,6 +556,12 @@
       "bytes": 4728422,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.6315553869990254,
+      "bytes": 5703905,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -98,3 +98,4 @@
 {"step": 852300, "loss": 0.5323869585990906, "lr": 3e-05, "grad_norm": 0.28288396236856206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124753180994, "tokens_per_second": 152524.21277255262, "data_wait_s": 1.3941998986410908e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 852400, "loss": 0.5360299348831177, "lr": 3e-05, "grad_norm": 0.3001661669295489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124767724603, "tokens_per_second": 136858.40141233857, "data_wait_s": 1.6341000446118414e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 852500, "loss": 0.5522243976593018, "lr": 3e-05, "grad_norm": 0.3072252878219012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124782212202, "tokens_per_second": 151555.68822367024, "data_wait_s": 1.205500120704528e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 852600, "loss": 0.5513781309127808, "lr": 3e-05, "grad_norm": 0.3217502851203123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124796715130, "tokens_per_second": 124554.46547856432, "data_wait_s": 1.6227999367401935e-05, "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:ebc05a6e67b0953a33c5420505ada0bd3c361cd8ee0a2b0753f88f69bf8923a3
-size 213531
+oid sha256:c1350a7bfd52d285b3cd92d17bba9b4d3b0584409091fa7413f85f23afac44f8
+size 215649