Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76190.00945474795,
-      "required_pages_per_second_upper_estimate": 0.0066039569177240125
+      "required_token_positions_per_second": 75992.09461447012,
+      "required_pages_per_second_upper_estimate": 0.0065868021609793055
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9523.751181843494,
-      "required_pages_per_second_upper_estimate": 0.0037991667392067554
+      "required_token_positions_per_second": 9499.011826808764,
+      "required_pages_per_second_upper_estimate": 0.0037892978405970816
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19047.50236368699,
-      "required_pages_per_second_upper_estimate": 0.0044063044472467255
+      "required_token_positions_per_second": 18998.02365361753,
+      "required_pages_per_second_upper_estimate": 0.004394858418467712
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28571.253545530482,
-      "required_pages_per_second_upper_estimate": 0.001739839439140473
+      "required_token_positions_per_second": 28497.035480426293,
+      "required_pages_per_second_upper_estimate": 0.0017353199483677217
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28571.253545530482,
-      "required_pages_per_second_upper_estimate": 0.0021994941882267543
+      "required_token_positions_per_second": 28497.035480426293,
+      "required_pages_per_second_upper_estimate": 0.0021937806761262806
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28571.253545530482,
-      "required_pages_per_second_upper_estimate": 0.354759347201044
+      "required_token_positions_per_second": 28497.035480426293,
+      "required_pages_per_second_upper_estimate": 0.35383780722433533
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1259711569531703,
-    "producer_p95_seconds": 0.3234117689971754,
-    "gpu_update_seconds": 0.9676859279534513,
-    "producer_batches_per_second": 7.938325122883085,
-    "gpu_batches_per_second": 1.0333931403910042,
-    "producer_headroom": 7.681805512933314,
+    "producer_batch_seconds": 0.10179227217977882,
+    "producer_p95_seconds": 0.3004091230031918,
+    "gpu_update_seconds": 0.9702061822883482,
+    "producer_batches_per_second": 9.823928463192821,
+    "gpu_batches_per_second": 1.0307087485686595,
+    "producer_headroom": 9.531236129348148,
     "sustainable": true,
-    "buffer_seconds": 60.96421346106743,
+    "buffer_seconds": 60.15278330187759,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 267267.9148146255,
+    "generation_bytes_per_second": 267208.466768932,
     "upload_bytes_per_second": 2304001.222910685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.349472840586415,
+    "estimated_full_shards_per_hour": 57.336716673860266,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -178,3 +178,4 @@
 {"step": 298000, "loss": 0.572965245693922, "lr": 0.0003, "grad_norm": 0.14221758515788127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43619684108, "tokens_per_second": 149950.25379936505, "data_wait_s": 1.3213000784162432e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 298100, "loss": 0.5779675506055355, "lr": 0.0003, "grad_norm": 0.15523465119748014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43634281008, "tokens_per_second": 148972.73237010153, "data_wait_s": 1.3013999705435708e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 298200, "loss": 0.5542546007782221, "lr": 0.0003, "grad_norm": 0.14852502627466904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43648809275, "tokens_per_second": 148614.5832521336, "data_wait_s": 1.6843998309923336e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 298300, "loss": 0.6011155806481838, "lr": 0.0003, "grad_norm": 0.15396914818704283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43663407844, "tokens_per_second": 149417.20726141307, "data_wait_s": 9.9300013971515e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b727365c2810cacfb23fc9c84227206026f8027c1a887689649682f409d36421
-size 383754
+oid sha256:22c97c219150a49253dfa2bf1a05acd6b005e40e47619cee952ab7b82e28c7cb
+size 385872