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": 76234.98475931358,
-      "required_pages_per_second_upper_estimate": 0.005274113135184577
+      "required_token_positions_per_second": 76909.86900180364,
+      "required_pages_per_second_upper_estimate": 0.005320803193027228
     },
     {
       "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": 9529.373094914197,
-      "required_pages_per_second_upper_estimate": 0.00398974951890177
+      "required_token_positions_per_second": 9613.733625225455,
+      "required_pages_per_second_upper_estimate": 0.004025069511294897
     },
     {
       "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": 19058.746189828395,
-      "required_pages_per_second_upper_estimate": 0.003969434218404973
+      "required_token_positions_per_second": 19227.46725045091,
+      "required_pages_per_second_upper_estimate": 0.00400457436585906
     },
     {
       "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": 28588.119284742595,
-      "required_pages_per_second_upper_estimate": 0.0017138140995896765
+      "required_token_positions_per_second": 28841.200875676364,
+      "required_pages_per_second_upper_estimate": 0.0017289859545329418
     },
     {
       "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": 28588.119284742595,
-      "required_pages_per_second_upper_estimate": 0.002222600528541234
+      "required_token_positions_per_second": 28841.200875676364,
+      "required_pages_per_second_upper_estimate": 0.0022422765090480616
     },
     {
       "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": 28588.119284742595,
-      "required_pages_per_second_upper_estimate": 0.3666976986537191
+      "required_token_positions_per_second": 28841.200875676364,
+      "required_pages_per_second_upper_estimate": 0.3699439575643766
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08655574700773627,
-    "producer_p95_seconds": 0.25243364100242616,
-    "gpu_update_seconds": 0.9671150356069651,
-    "producer_batches_per_second": 11.55324787284917,
-    "gpu_batches_per_second": 1.0340031570002386,
-    "producer_headroom": 11.17331972792662,
+    "producer_batch_seconds": 0.2045562653981392,
+    "producer_p95_seconds": 0.2599544250006147,
+    "gpu_update_seconds": 0.9586285993839228,
+    "producer_batches_per_second": 4.888630509819118,
+    "gpu_batches_per_second": 1.043156860376026,
+    "producer_headroom": 4.686381018533413,
     "sustainable": true,
-    "buffer_seconds": 60.9282472432388,
+    "buffer_seconds": 60.393601761187135,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 173,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 573812.6548489522,
+    "generation_bytes_per_second": 551698.9594829893,
     "upload_bytes_per_second": 5613278.311254591,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.563419027812124,
+    "estimated_full_shards_per_hour": 118.38175381057032,
     "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": [

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 505400, "loss": 0.6163082122802734, "lr": 0.0003, "grad_norm": 0.1687311998915468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73973636957, "tokens_per_second": 153581.08781746763, "data_wait_s": 1.2981003237655386e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 505500, "loss": 0.5667534470558167, "lr": 0.0003, "grad_norm": 0.17304352353442695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73988462488, "tokens_per_second": 154329.1296789565, "data_wait_s": 1.0910000128205866e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 505600, "loss": 0.6149583458900452, "lr": 0.0003, "grad_norm": 0.17727348962128936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74003082842, "tokens_per_second": 152165.66491905748, "data_wait_s": 1.499199788668193e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 505700, "loss": 0.6090046167373657, "lr": 0.0003, "grad_norm": 0.15855559229407937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74017608939, "tokens_per_second": 151303.1249323467, "data_wait_s": 9.109000529861078e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e5ccba050504e0b3f937e9489eb9969d39190d9227da4eff21e35903117a5ed
-size 439605
+oid sha256:6ec5f27f9f7f4310aa6576380007ef64e665a6de31aff39f983761659bac2898
+size 441723