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": 74333.58647398642,
-      "required_pages_per_second_upper_estimate": 0.005142569989953788
+      "required_token_positions_per_second": 77116.80661270273,
+      "required_pages_per_second_upper_estimate": 0.005335119617110637
     },
     {
       "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": 9291.698309248302,
-      "required_pages_per_second_upper_estimate": 0.0038902400493573703
+      "required_token_positions_per_second": 9639.60082658784,
+      "required_pages_per_second_upper_estimate": 0.004035899568336739
     },
     {
       "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": 18583.396618496605,
-      "required_pages_per_second_upper_estimate": 0.0038704314385077537
+      "required_token_positions_per_second": 19279.20165317568,
+      "required_pages_per_second_upper_estimate": 0.004015349277618683
     },
     {
       "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": 27875.09492774491,
-      "required_pages_per_second_upper_estimate": 0.0016710693781128191
+      "required_token_positions_per_second": 28918.802479763523,
+      "required_pages_per_second_upper_estimate": 0.0017336380522071783
     },
     {
       "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": 27875.09492774491,
-      "required_pages_per_second_upper_estimate": 0.00216716602104736
+      "required_token_positions_per_second": 28918.802479763523,
+      "required_pages_per_second_upper_estimate": 0.0022483096924324547
     },
     {
       "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": 27875.09492774491,
-      "required_pages_per_second_upper_estimate": 0.35755178778805957
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28918.802479763523,
+      "required_pages_per_second_upper_estimate": 0.37093934761949593
     }
   ],
   "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": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06229197961722832,
-    "producer_p95_seconds": 0.21709130999988702,
-    "gpu_update_seconds": 0.9918531245065331,
-    "producer_batches_per_second": 16.053431053962303,
-    "gpu_batches_per_second": 1.008213792236144,
-    "producer_headroom": 15.922645749922717,
+    "producer_batch_seconds": 0.08553106301565094,
+    "producer_p95_seconds": 0.20938527499856718,
+    "gpu_update_seconds": 0.956056185913895,
+    "producer_batches_per_second": 11.691658734757155,
+    "gpu_batches_per_second": 1.0459636313571876,
+    "producer_headroom": 11.1778826569588,
     "sustainable": true,
-    "buffer_seconds": 60.503040594898515,
+    "buffer_seconds": 60.23153971257538,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305010.00269171305,
-    "upload_bytes_per_second": 5478732.163127649,
+    "generation_bytes_per_second": 340442.04527818813,
+    "upload_bytes_per_second": 5376747.698464667,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.44804630816978,
+    "estimated_full_shards_per_hour": 73.05093783148988,
     "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": 1.2060352290000083,
-      "bytes": 11046742,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.0752750450001258,
       "bytes": 6242403,
@@ -562,6 +556,12 @@
       "bytes": 6359822,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1105975430000399,
+      "bytes": 2661454,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -132,3 +132,4 @@
 {"step": 410300, "loss": 0.5704032182693481, "lr": 0.0003, "grad_norm": 0.16303334227756025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60057065977, "tokens_per_second": 152063.24380120283, "data_wait_s": 1.6596999557805248e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 410400, "loss": 0.5478192567825317, "lr": 0.0003, "grad_norm": 0.14434415254926053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60071842138, "tokens_per_second": 153785.14796797503, "data_wait_s": 1.5442999938386492e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 410500, "loss": 0.5681799054145813, "lr": 0.0003, "grad_norm": 0.1623207670027057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60086465650, "tokens_per_second": 150977.93994714442, "data_wait_s": 2.4000999474083073e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 410600, "loss": 0.6016080975532532, "lr": 0.0003, "grad_norm": 0.1727383479707727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60101196151, "tokens_per_second": 152161.6338070798, "data_wait_s": 1.414700091117993e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ce779a4e6746c87b58810b09094b3700f8dd08e4a9f850999a525dfde0118bf
-size 286326
+oid sha256:72e866e5028e32a578ea3f99b9c1202fa1971a659949e1639a8ac177f17db35b
+size 288444