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": 77043.6053918256,
-      "required_pages_per_second_upper_estimate": 0.005330055386800123
+      "required_token_positions_per_second": 76782.39731124735,
+      "required_pages_per_second_upper_estimate": 0.005311984405699471
     },
     {
       "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": 9630.4506739782,
-      "required_pages_per_second_upper_estimate": 0.004032068590515996
+      "required_token_positions_per_second": 9597.799663905918,
+      "required_pages_per_second_upper_estimate": 0.004018398294429356
     },
     {
       "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": 19260.9013479564,
-      "required_pages_per_second_upper_estimate": 0.004011537806653009
+      "required_token_positions_per_second": 19195.599327811837,
+      "required_pages_per_second_upper_estimate": 0.003997937117986976
     },
     {
       "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": 28891.352021934596,
-      "required_pages_per_second_upper_estimate": 0.0017319924391747561
+      "required_token_positions_per_second": 28793.398991717757,
+      "required_pages_per_second_upper_estimate": 0.0017261203045788739
     },
     {
       "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": 28891.352021934596,
-      "required_pages_per_second_upper_estimate": 0.0022461755400780595
+      "required_token_positions_per_second": 28793.398991717757,
+      "required_pages_per_second_upper_estimate": 0.0022385601228285465
     },
     {
       "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": 28891.352021934596,
-      "required_pages_per_second_upper_estimate": 0.3705872426204717
+      "required_token_positions_per_second": 28793.398991717757,
+      "required_pages_per_second_upper_estimate": 0.3693308063226197
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.06324522479673078,
-    "producer_p95_seconds": 0.2149647959995491,
-    "gpu_update_seconds": 0.956964560848844,
-    "producer_batches_per_second": 15.811470402927418,
-    "gpu_batches_per_second": 1.044970776256315,
-    "producer_headroom": 15.13101683051193,
+    "producer_batch_seconds": 0.11713293551551374,
+    "producer_p95_seconds": 0.23200444800022524,
+    "gpu_update_seconds": 0.960220084053042,
+    "producer_batches_per_second": 8.537308448720253,
+    "gpu_batches_per_second": 1.0414279149203471,
+    "producer_headroom": 8.197695036216908,
     "sustainable": true,
-    "buffer_seconds": 60.28876733347717,
+    "buffer_seconds": 60.49386529534165,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330630.9173485007,
+    "generation_bytes_per_second": 355204.2613708904,
     "upload_bytes_per_second": 5176293.278542278,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.94569816914812,
+    "estimated_full_shards_per_hour": 76.21856575818094,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -201,3 +201,4 @@
 {"step": 624500, "loss": 0.5853891968727112, "lr": 0.0001, "grad_norm": 0.18553739924719578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91410018711, "tokens_per_second": 151223.35202673485, "data_wait_s": 1.61259995365981e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 624600, "loss": 0.5893364548683167, "lr": 0.0001, "grad_norm": 0.15714717350095056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91424676083, "tokens_per_second": 151960.73589837409, "data_wait_s": 1.584900019224733e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 624700, "loss": 0.6741824150085449, "lr": 0.0001, "grad_norm": 0.21393643831026557, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91439406413, "tokens_per_second": 153202.57430646956, "data_wait_s": 1.302200325881131e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 624800, "loss": 0.5320255756378174, "lr": 0.0001, "grad_norm": 0.1538710211550954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91454166444, "tokens_per_second": 153312.61372443134, "data_wait_s": 1.692800287855789e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15f97405f67e571d8617a3fc1d67a11821ccd04ebf674b868d34198dd5381485
-size 433251
+oid sha256:f23e25c855b953459f9efaacc9aa786fca98448f86a998e30cb514464c0d52e9
+size 435369