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": 76717.42552495262,
-      "required_pages_per_second_upper_estimate": 0.005307489506768287
+      "required_token_positions_per_second": 76492.8407228775,
+      "required_pages_per_second_upper_estimate": 0.005291952209052196
     },
     {
       "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": 9589.678190619077,
-      "required_pages_per_second_upper_estimate": 0.004014998003159804
+      "required_token_positions_per_second": 9561.605090359688,
+      "required_pages_per_second_upper_estimate": 0.004003244382314194
     },
     {
       "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": 19179.356381238154,
-      "required_pages_per_second_upper_estimate": 0.003994554140570985
+      "required_token_positions_per_second": 19123.210180719376,
+      "required_pages_per_second_upper_estimate": 0.003982860367677803
     },
     {
       "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": 28769.03457185723,
-      "required_pages_per_second_upper_estimate": 0.0017246596948105504
+      "required_token_positions_per_second": 28684.815271079064,
+      "required_pages_per_second_upper_estimate": 0.0017196108763243255
     },
     {
       "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": 28769.03457185723,
-      "required_pages_per_second_upper_estimate": 0.0022366658963521475
+      "required_token_positions_per_second": 28684.815271079064,
+      "required_pages_per_second_upper_estimate": 0.0022301182161582
     },
     {
       "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": 28769.03457185723,
-      "required_pages_per_second_upper_estimate": 0.36901828570512474
+      "required_token_positions_per_second": 28684.815271079064,
+      "required_pages_per_second_upper_estimate": 0.36793801094238227
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2607842184688991,
-    "producer_p95_seconds": 0.26586338299966883,
-    "gpu_update_seconds": 0.96103329192166,
-    "producer_batches_per_second": 3.834587866823924,
-    "gpu_batches_per_second": 1.0405466786696047,
-    "producer_headroom": 3.6851666008166517,
+    "producer_batch_seconds": 0.12345343640626538,
+    "producer_p95_seconds": 0.2665688930028409,
+    "gpu_update_seconds": 0.963854908554199,
+    "producer_batches_per_second": 8.10022004336243,
+    "gpu_batches_per_second": 1.0375005523393759,
+    "producer_headroom": 7.8074368491639845,
     "sustainable": true,
-    "buffer_seconds": 60.54509739106458,
+    "buffer_seconds": 60.72285923891454,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451285.25893055904,
-    "upload_bytes_per_second": 3932602.734218998,
+    "generation_bytes_per_second": 467380.42680399393,
+    "upload_bytes_per_second": 3889459.3215677463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.83531118333414,
+    "estimated_full_shards_per_hour": 100.28895953264106,
     "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": 1.048425811000925,
-      "bytes": 5676510,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 33.502104078999764,
       "bytes": 11160724,
@@ -562,6 +556,12 @@
       "bytes": 16563668,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.482665261002694,
+      "bytes": 2527538,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -237,3 +237,4 @@
 {"step": 668700, "loss": 0.52390456199646, "lr": 0.0001, "grad_norm": 0.19498226587115416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97880620963, "tokens_per_second": 153817.60367502773, "data_wait_s": 1.3876000593882054e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 668800, "loss": 0.5165662169456482, "lr": 0.0001, "grad_norm": 0.1785276674904329, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97895343309, "tokens_per_second": 152477.83515864637, "data_wait_s": 1.4309000107459724e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 668900, "loss": 0.6067279577255249, "lr": 0.0001, "grad_norm": 0.21197738807438912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97910002144, "tokens_per_second": 152460.0939288567, "data_wait_s": 5.559999408433214e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 669000, "loss": 0.49687978625297546, "lr": 0.0001, "grad_norm": 0.18339128986633188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97924628773, "tokens_per_second": 151943.89155345468, "data_wait_s": 1.3961998774902895e-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 sha256:7e3b758eadf2833ad5623eddcbc5fe318fb717013cd5c92587967c81219c3382
-size 509499
+oid sha256:1af1f964205998a053ff63fe1e8be3b2c209a0f2b505052d779f121ed702e65d
+size 511617