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": 76059.81374876449,
-      "required_pages_per_second_upper_estimate": 0.005261994398222048
+      "required_token_positions_per_second": 76925.40144986997,
+      "required_pages_per_second_upper_estimate": 0.005321877763824701
     },
     {
       "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": 9507.47671859556,
-      "required_pages_per_second_upper_estimate": 0.00398058196338549
+      "required_token_positions_per_second": 9615.675181233746,
+      "required_pages_per_second_upper_estimate": 0.0040258824002512685
     },
     {
       "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": 19014.95343719112,
-      "required_pages_per_second_upper_estimate": 0.003960313342923179
+      "required_token_positions_per_second": 19231.350362467492,
+      "required_pages_per_second_upper_estimate": 0.004005383115687564
     },
     {
       "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": 28522.430155786682,
-      "required_pages_per_second_upper_estimate": 0.0017098761366102617
+      "required_token_positions_per_second": 28847.02554370124,
+      "required_pages_per_second_upper_estimate": 0.0017293351344872798
     },
     {
       "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": 28522.430155786682,
-      "required_pages_per_second_upper_estimate": 0.002217493487700854
+      "required_token_positions_per_second": 28847.02554370124,
+      "required_pages_per_second_upper_estimate": 0.002242729351366988
     },
     {
       "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": 28522.430155786682,
-      "required_pages_per_second_upper_estimate": 0.3658551090389641
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28847.02554370124,
+      "required_pages_per_second_upper_estimate": 0.37001867015175843
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5090801387187298,
-    "producer_p95_seconds": 0.2816256900005101,
-    "gpu_update_seconds": 0.9693423684093315,
-    "producer_batches_per_second": 1.9643272717667477,
-    "gpu_batches_per_second": 1.0316272481114974,
-    "producer_headroom": 1.9041056499454199,
+    "producer_batch_seconds": 0.09168338842937374,
+    "producer_p95_seconds": 0.27760642200155417,
+    "gpu_update_seconds": 0.9584350371969963,
+    "producer_batches_per_second": 10.907101244085538,
+    "gpu_batches_per_second": 1.0433675326859535,
+    "producer_headroom": 10.453747986586528,
     "sustainable": true,
-    "buffer_seconds": 60.09922684137855,
+    "buffer_seconds": 60.381407343410764,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337595.49289931636,
-    "upload_bytes_per_second": 4878950.727388946,
+    "generation_bytes_per_second": 339377.67936794454,
+    "upload_bytes_per_second": 4930189.1618287815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.44013395533197,
+    "estimated_full_shards_per_hour": 72.8225496843219,
     "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": 2.129564337999909,
-      "bytes": 5884538,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.1728028199995606,
       "bytes": 6147166,
@@ -562,6 +556,12 @@
       "bytes": 11926092,
       "pages": 29,
       "shards": 4
+    },
+    {
+      "seconds": 1.4296780900003796,
+      "bytes": 6878310,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -236,3 +236,4 @@
 {"step": 866100, "loss": 0.49722954630851746, "lr": 3e-05, "grad_norm": 0.31966955954025744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126772931548, "tokens_per_second": 151866.74268190228, "data_wait_s": 6.505000783363357e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 866200, "loss": 0.5219362378120422, "lr": 3e-05, "grad_norm": 0.3036566328829848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126787579553, "tokens_per_second": 145926.98003303452, "data_wait_s": 1.4743000065209344e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 866300, "loss": 0.5333990454673767, "lr": 3e-05, "grad_norm": 0.2994699033609311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126802122267, "tokens_per_second": 150933.3663034294, "data_wait_s": 1.697700281511061e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 866400, "loss": 0.5334560871124268, "lr": 3e-05, "grad_norm": 0.3110045366919706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126816932540, "tokens_per_second": 153359.14419664512, "data_wait_s": 1.0814997949637473e-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:e01fee181b4c2e38c9245e82a2dfe1db90da237acc2b35dcdfe0ddd2581e6514
-size 508164
+oid sha256:142f586a02d01c276b3cdbbe168a1d13d2eacef2c3e32be383deb9f803b985e2
+size 510282