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": 77258.42315392195,
-      "required_pages_per_second_upper_estimate": 0.00534491697802783
+      "required_token_positions_per_second": 77176.89841787542,
+      "required_pages_per_second_upper_estimate": 0.005339276907624682
     },
     {
       "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": 9657.302894240243,
-      "required_pages_per_second_upper_estimate": 0.004043311054401592
+      "required_token_positions_per_second": 9647.112302234427,
+      "required_pages_per_second_upper_estimate": 0.004039044466332516
     },
     {
       "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": 19314.605788480487,
-      "required_pages_per_second_upper_estimate": 0.00402272302533284
+      "required_token_positions_per_second": 19294.224604468855,
+      "required_pages_per_second_upper_estimate": 0.004018478162191183
     },
     {
       "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": 28971.908682720732,
-      "required_pages_per_second_upper_estimate": 0.0017368216879859863
+      "required_token_positions_per_second": 28941.33690670328,
+      "required_pages_per_second_upper_estimate": 0.001734988956694139
     },
     {
       "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": 28971.908682720732,
-      "required_pages_per_second_upper_estimate": 0.0022524384661228775
+      "required_token_positions_per_second": 28941.33690670328,
+      "required_pages_per_second_upper_estimate": 0.002250061645008552
     },
     {
       "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": 28971.908682720732,
-      "required_pages_per_second_upper_estimate": 0.3716205369700329
+      "required_token_positions_per_second": 28941.33690670328,
+      "required_pages_per_second_upper_estimate": 0.3712283950526966
     }
   ],
   "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": 0.09933180085170079,
-    "producer_p95_seconds": 0.29079541199826053,
-    "gpu_update_seconds": 0.9543037120122386,
-    "producer_batches_per_second": 10.067269408444211,
-    "gpu_batches_per_second": 1.0478844286284987,
-    "producer_headroom": 9.607232566305564,
+    "producer_batch_seconds": 0.09608320560147376,
+    "producer_p95_seconds": 0.27539721799985273,
+    "gpu_update_seconds": 0.9553117773766794,
+    "producer_batches_per_second": 10.407646099441353,
+    "gpu_batches_per_second": 1.0467786786278674,
+    "producer_headroom": 9.942546893564783,
     "sustainable": true,
-    "buffer_seconds": 60.12113385677103,
+    "buffer_seconds": 60.1846419747308,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307726.94420068036,
-    "upload_bytes_per_second": 4765741.759505369,
+    "generation_bytes_per_second": 397695.1736651185,
+    "upload_bytes_per_second": 4823252.873992993,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.03103870883282,
+    "estimated_full_shards_per_hour": 85.33612639870803,
     "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": 1.2694779230005224,
-      "bytes": 2738639,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.369231013002718,
       "bytes": 10621268,
@@ -562,6 +556,12 @@
       "bytes": 2657150,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 0.9289632090003579,
+      "bytes": 5852982,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -297,3 +297,4 @@
 {"step": 714700, "loss": 0.5153366327285767, "lr": 8e-05, "grad_norm": 0.2069151599215396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104612159098, "tokens_per_second": 151318.54636267663, "data_wait_s": 1.5069999790284783e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 714800, "loss": 0.552908182144165, "lr": 8e-05, "grad_norm": 0.2087740568057489, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104626872512, "tokens_per_second": 153083.12809174953, "data_wait_s": 9.734001650940627e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 714900, "loss": 0.5387517213821411, "lr": 8e-05, "grad_norm": 0.21496947351464413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104641426324, "tokens_per_second": 151266.63435850193, "data_wait_s": 1.8170001567341387e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 715000, "loss": 0.49898889660835266, "lr": 8e-05, "grad_norm": 0.23847586120249772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104656097809, "tokens_per_second": 152420.1536742034, "data_wait_s": 1.3430999388219789e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:663f67839ba2b99c6f639c34a4f0904ccb34aa4b86abbb0e508e8ab886d5b1a7
-size 637362
+oid sha256:3e110aaff4a831989ce3119ff5db8a9c77e10ba35c25e062cc244cb3101f1ab1
+size 639480