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": 75802.31052653243,
-      "required_pages_per_second_upper_estimate": 0.00524417972781824
+      "required_token_positions_per_second": 76316.70706820943,
+      "required_pages_per_second_upper_estimate": 0.005279766874135874
     },
     {
       "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": 9475.288815816553,
-      "required_pages_per_second_upper_estimate": 0.003967105560651763
+      "required_token_positions_per_second": 9539.58838352618,
+      "required_pages_per_second_upper_estimate": 0.003994026446924123
     },
     {
       "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": 18950.577631633107,
-      "required_pages_per_second_upper_estimate": 0.003946905560329603
+      "required_token_positions_per_second": 19079.17676705236,
+      "required_pages_per_second_upper_estimate": 0.003973689368850171
     },
     {
       "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": 28425.86644744966,
-      "required_pages_per_second_upper_estimate": 0.0017040872897397027
+      "required_token_positions_per_second": 28618.76515057854,
+      "required_pages_per_second_upper_estimate": 0.0017156512724530136
     },
     {
       "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": 28425.86644744966,
-      "required_pages_per_second_upper_estimate": 0.0022099860841165164
+      "required_token_positions_per_second": 28618.76515057854,
+      "required_pages_per_second_upper_estimate": 0.002224983109813061
     },
     {
       "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": 28425.86644744966,
-      "required_pages_per_second_upper_estimate": 0.36461649347044883
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28618.76515057854,
+      "required_pages_per_second_upper_estimate": 0.3670907909156955
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.06326813192967595,
-    "producer_p95_seconds": 0.127287774000024,
-    "gpu_update_seconds": 0.9726352599000742,
-    "producer_batches_per_second": 15.805745633702667,
-    "gpu_batches_per_second": 1.0281346371328726,
-    "producer_headroom": 15.373225512350857,
+    "producer_batch_seconds": 0.10251773209374804,
+    "producer_p95_seconds": 0.1337981230000196,
+    "gpu_update_seconds": 0.9660794186796381,
+    "producer_batches_per_second": 9.754410086691571,
+    "gpu_batches_per_second": 1.0351115867541427,
+    "producer_headroom": 9.42353482611379,
     "sustainable": true,
-    "buffer_seconds": 60.3033861138046,
+    "buffer_seconds": 60.8630033768172,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 103,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 836892.4980165607,
+    "generation_bytes_per_second": 597444.4173980614,
     "upload_bytes_per_second": 285529.7561989582,
     "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.78882410703953,
+    "estimated_full_shards_per_hour": 64.09883208969298,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -1,3 +1,4 @@
 {"step": 842600, "loss": 0.5411972999572754, "lr": 3e-05, "grad_norm": 0.28530642342954515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123332630293, "tokens_per_second": 152128.42886191164, "data_wait_s": 1.4998999972704041e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 842700, "loss": 0.5061265230178833, "lr": 3e-05, "grad_norm": 0.2941202744714354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123347418029, "tokens_per_second": 153469.918480673, "data_wait_s": 1.93109999599983e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 842800, "loss": 0.4928826093673706, "lr": 3e-05, "grad_norm": 0.29498160693257575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123362142415, "tokens_per_second": 152243.4566548345, "data_wait_s": 1.890800001547177e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 842900, "loss": 0.543428897857666, "lr": 3e-05, "grad_norm": 0.2963097915073355, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123376940569, "tokens_per_second": 153854.7979767004, "data_wait_s": 1.0838999969564611e-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 sha256:35d35ddd7c4766a79a2f419e2e84caa5256ac5f4f585b05f7141ee54a3f41441
-size 6519
+oid sha256:b8edb012784f845d7e29d16543f8b47b6af40b28766586d1303cd9a9809d61b0
+size 8637