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": 77203.30671405637,
-      "required_pages_per_second_upper_estimate": 0.005341103894830169
+      "required_token_positions_per_second": 77287.36075835252,
+      "required_pages_per_second_upper_estimate": 0.0053469189486468295
     },
     {
       "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": 9650.413339257046,
-      "required_pages_per_second_upper_estimate": 0.004040426541600395
+      "required_token_positions_per_second": 9660.920094794064,
+      "required_pages_per_second_upper_estimate": 0.00404482550073774
     },
     {
       "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": 19300.82667851409,
-      "required_pages_per_second_upper_estimate": 0.004019853200106405
+      "required_token_positions_per_second": 19321.84018958813,
+      "required_pages_per_second_upper_estimate": 0.004024229760299574
     },
     {
       "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": 28951.240017771135,
-      "required_pages_per_second_upper_estimate": 0.0017355826330814818
+      "required_token_positions_per_second": 28982.760284382195,
+      "required_pages_per_second_upper_estimate": 0.0017374722249361547
     },
     {
       "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": 28951.240017771135,
-      "required_pages_per_second_upper_estimate": 0.0022508315683348957
+      "required_token_positions_per_second": 28982.760284382195,
+      "required_pages_per_second_upper_estimate": 0.002253282130420905
     },
     {
       "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": 28951.240017771135,
-      "required_pages_per_second_upper_estimate": 0.37135542152834283
+      "required_token_positions_per_second": 28982.760284382195,
+      "required_pages_per_second_upper_estimate": 0.371759729664604
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10230875125780514,
-    "producer_p95_seconds": 0.14922776200000953,
-    "gpu_update_seconds": 0.9549850017833029,
-    "producer_batches_per_second": 9.77433491960161,
-    "gpu_batches_per_second": 1.0471368640686898,
-    "producer_headroom": 9.334343250626343,
+    "producer_batch_seconds": 0.06581913967185571,
+    "producer_p95_seconds": 0.1402002760000869,
+    "gpu_update_seconds": 0.9539464056809851,
+    "producer_batches_per_second": 15.1931490594612,
+    "gpu_batches_per_second": 1.0482769200080366,
+    "producer_headroom": 14.493449936248453,
     "sustainable": true,
-    "buffer_seconds": 60.16405511234809,
+    "buffer_seconds": 60.09862355790206,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367332.59541420545,
+    "generation_bytes_per_second": 383042.87084597046,
     "upload_bytes_per_second": 4994445.429721233,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.8210239107096,
+    "estimated_full_shards_per_hour": 82.19208330187163,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -17,3 +17,4 @@
 {"step": 558800, "loss": 0.5554922223091125, "lr": 0.0003, "grad_norm": 0.1730167379325697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81790350258, "tokens_per_second": 152810.90438993459, "data_wait_s": 1.1388000075385207e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 558900, "loss": 0.5704553127288818, "lr": 0.0003, "grad_norm": 0.14818424246611112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81805104411, "tokens_per_second": 153543.70651069615, "data_wait_s": 1.4848000319034327e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 559000, "loss": 0.6242825388908386, "lr": 0.0003, "grad_norm": 0.1672461842499151, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81819922760, "tokens_per_second": 154236.35030483408, "data_wait_s": 1.2404000244714553e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 559100, "loss": 0.5916821956634521, "lr": 0.0003, "grad_norm": 0.20054137954538745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81834664568, "tokens_per_second": 153743.3016488649, "data_wait_s": 1.3612999737233622e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64b017de91d583db38dd124d5b50864c7e1ad9b0082b1c3164897717beeb59e3
-size 40407
+oid sha256:1ed9b6629ba6e423fa4053f7e7e548aa3d2ec9e433cade5dc123a0ba420b789b
+size 42525