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": 77602.28811995091,
-      "required_pages_per_second_upper_estimate": 0.005368706354254361
+      "required_token_positions_per_second": 77195.16895574742,
+      "required_pages_per_second_upper_estimate": 0.005340540905828155
     },
     {
       "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": 9700.286014993864,
-      "required_pages_per_second_upper_estimate": 0.004061307189471503
+      "required_token_positions_per_second": 9649.396119468427,
+      "required_pages_per_second_upper_estimate": 0.004040000652916865
     },
     {
       "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": 19400.57202998773,
-      "required_pages_per_second_upper_estimate": 0.0040406275263565545
+      "required_token_positions_per_second": 19298.792238936854,
+      "required_pages_per_second_upper_estimate": 0.004019429479994246
     },
     {
       "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": 29100.858044981593,
-      "required_pages_per_second_upper_estimate": 0.0017445520053592492
+      "required_token_positions_per_second": 28948.18835840528,
+      "required_pages_per_second_upper_estimate": 0.001735399690762113
     },
     {
       "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": 29100.858044981593,
-      "required_pages_per_second_upper_estimate": 0.002262463711850357
+      "required_token_positions_per_second": 28948.18835840528,
+      "required_pages_per_second_upper_estimate": 0.0022505943152420316
     },
     {
       "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": 29100.858044981593,
-      "required_pages_per_second_upper_estimate": 0.3732745609340772
+      "required_token_positions_per_second": 28948.18835840528,
+      "required_pages_per_second_upper_estimate": 0.3713162781186142
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.056567684632796045,
-    "producer_p95_seconds": 0.1767701959997794,
-    "gpu_update_seconds": 0.9500750787919762,
-    "producer_batches_per_second": 17.677937615644137,
-    "gpu_batches_per_second": 1.0525483957241606,
-    "producer_headroom": 16.795367973062742,
+    "producer_batch_seconds": 0.3353464534454389,
+    "producer_p95_seconds": 0.20689823600150703,
+    "gpu_update_seconds": 0.95508567436733,
+    "producer_batches_per_second": 2.981990683741347,
+    "gpu_batches_per_second": 1.0470264886575984,
+    "producer_headroom": 2.8480565831381996,
     "sustainable": true,
-    "buffer_seconds": 60.804805042686475,
+    "buffer_seconds": 60.17039748514179,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350255.4186814308,
-    "upload_bytes_per_second": 3224443.4778573196,
+    "generation_bytes_per_second": 322449.68271191884,
+    "upload_bytes_per_second": 3249130.2766539133,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.15665932018464,
+    "estimated_full_shards_per_hour": 69.19019566553281,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 2379174,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1466768219997903,
+      "bytes": 6100233,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 654000, "loss": 0.5774896740913391, "lr": 0.0001, "grad_norm": 0.19368160866251724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95728641982, "tokens_per_second": 153102.53756921992, "data_wait_s": 1.619199974811636e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 654100, "loss": 0.5593799948692322, "lr": 0.0001, "grad_norm": 0.19241948478255352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95743377997, "tokens_per_second": 153748.94903408128, "data_wait_s": 9.348001185571775e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 654200, "loss": 0.53191739320755, "lr": 0.0001, "grad_norm": 0.1870930580141497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95758071937, "tokens_per_second": 153250.9483127806, "data_wait_s": 1.4536000890075229e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 654300, "loss": 0.5371429920196533, "lr": 0.0001, "grad_norm": 0.1857088191072252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95772680477, "tokens_per_second": 152713.97499857313, "data_wait_s": 1.0858999303309247e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a1a308aa2a6b11c7d1e1507b16bc07c34234f4a2a8625db363af4c2d59db0ed2
-size 195804
+oid sha256:87c19fa72668ad62da61c12f964b362a4244241ac23868cde8a7612a22c44c8e
+size 197922