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": 75711.298881415,
-      "required_pages_per_second_upper_estimate": 0.0052378833310329055
+      "required_token_positions_per_second": 77114.06681364063,
+      "required_pages_per_second_upper_estimate": 0.005334930071454311
     },
     {
       "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": 9463.912360176875,
-      "required_pages_per_second_upper_estimate": 0.003962342476242838
+      "required_token_positions_per_second": 9639.258351705079,
+      "required_pages_per_second_upper_estimate": 0.004035756181255015
     },
     {
       "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": 18927.82472035375,
-      "required_pages_per_second_upper_estimate": 0.003942166728944734
+      "required_token_positions_per_second": 19278.516703410158,
+      "required_pages_per_second_upper_estimate": 0.004015206620645864
     },
     {
       "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": 28391.737080530624,
-      "required_pages_per_second_upper_estimate": 0.0017020412863054321
+      "required_token_positions_per_second": 28917.77505511524,
+      "required_pages_per_second_upper_estimate": 0.0017335764596683003
     },
     {
       "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": 28391.737080530624,
-      "required_pages_per_second_upper_estimate": 0.002207332675957781
+      "required_token_positions_per_second": 28917.77505511524,
+      "required_pages_per_second_upper_estimate": 0.0022482298146852143
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28391.737080530624,
-      "required_pages_per_second_upper_estimate": 0.3641787185968705
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28917.77505511524,
+      "required_pages_per_second_upper_estimate": 0.3709261689192704
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13520461964066044,
-    "producer_p95_seconds": 0.2855433209988405,
-    "gpu_update_seconds": 0.9738044530906622,
-    "producer_batches_per_second": 7.396196983932548,
-    "gpu_batches_per_second": 1.0269002126928033,
-    "producer_headroom": 7.202449558889239,
+    "producer_batch_seconds": 0.0727084036641088,
+    "producer_p95_seconds": 0.2758577839995269,
+    "gpu_update_seconds": 0.9560901538000369,
+    "producer_batches_per_second": 13.753568358063568,
+    "gpu_batches_per_second": 1.0459264704541102,
+    "producer_headroom": 13.149651286760319,
     "sustainable": true,
-    "buffer_seconds": 60.37587609162106,
+    "buffer_seconds": 60.23367968940233,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358001.55127571075,
-    "upload_bytes_per_second": 3629836.3380909152,
+    "generation_bytes_per_second": 459538.0392612361,
+    "upload_bytes_per_second": 4956330.923610472,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.81879905417911,
+    "estimated_full_shards_per_hour": 98.60616572740375,
     "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": 1.716705255999841,
-      "bytes": 6235687,
-      "pages": 36,
-      "shards": 2
-    },
-    {
-      "seconds": 33.88602663399979,
-      "bytes": 7355362,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.5573600560001069,
       "bytes": 11648033,
@@ -562,6 +550,18 @@
       "bytes": 2460437,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.3531316250009695,
+      "bytes": 6148520,
+      "pages": 36,
+      "shards": 2
+    },
+    {
+      "seconds": 0.9429252529989753,
+      "bytes": 341818,
+      "pages": 9,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -223,3 +223,4 @@
 {"step": 864800, "loss": 0.5435219407081604, "lr": 3e-05, "grad_norm": 0.32043555136016144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126582788786, "tokens_per_second": 152000.72178230458, "data_wait_s": 1.5872999938437715e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 864900, "loss": 0.5208855271339417, "lr": 3e-05, "grad_norm": 0.2896373174278658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126597576438, "tokens_per_second": 153202.88513201484, "data_wait_s": 1.9161001546308398e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 865000, "loss": 0.49610868096351624, "lr": 3e-05, "grad_norm": 0.30979349364113046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126612249303, "tokens_per_second": 151404.03135483648, "data_wait_s": 1.536100171506405e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 865100, "loss": 0.514300525188446, "lr": 3e-05, "grad_norm": 0.304575595627324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126626759715, "tokens_per_second": 140371.71144499368, "data_wait_s": 8.618997526355088e-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 sha256:5a2aa3a9ce372bd74d5429b2a79ee392adee5ba6b2682fe4d86846fafac1a927
-size 480630
+oid sha256:7392428c3592cfd7451c038652e785d694d607a499b055408090eae983a4ec02
+size 482748