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": 77375.35979194005,
-      "required_pages_per_second_upper_estimate": 0.005353006925976304
+      "required_token_positions_per_second": 77094.46798309748,
+      "required_pages_per_second_upper_estimate": 0.005333574178881784
     },
     {
       "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": 9671.919973992506,
-      "required_pages_per_second_upper_estimate": 0.004049430920454529
+      "required_token_positions_per_second": 9636.808497887185,
+      "required_pages_per_second_upper_estimate": 0.004034730478620228
     },
     {
       "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": 19343.839947985012,
-      "required_pages_per_second_upper_estimate": 0.004028811729801001
+      "required_token_positions_per_second": 19273.61699577437,
+      "required_pages_per_second_upper_estimate": 0.004014186140759309
     },
     {
       "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": 29015.759921977522,
-      "required_pages_per_second_upper_estimate": 0.0017394505028224653
+      "required_token_positions_per_second": 28910.42549366155,
+      "required_pages_per_second_upper_estimate": 0.001733135864681282
     },
     {
       "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": 29015.759921977522,
-      "required_pages_per_second_upper_estimate": 0.0022558477070938757
+      "required_token_positions_per_second": 28910.42549366155,
+      "required_pages_per_second_upper_estimate": 0.0022476584186094943
     },
     {
       "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": 29015.759921977522,
-      "required_pages_per_second_upper_estimate": 0.3721830135834266
+      "required_token_positions_per_second": 28910.42549366155,
+      "required_pages_per_second_upper_estimate": 0.37083189663628674
     }
   ],
   "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": 0.05444897336717247,
-    "producer_p95_seconds": 0.18370804699952714,
-    "gpu_update_seconds": 0.9528614819789182,
-    "producer_batches_per_second": 18.36581919105392,
-    "gpu_batches_per_second": 1.0494704832891175,
-    "producer_headroom": 17.500081692144494,
+    "producer_batch_seconds": 0.0864632906796885,
+    "producer_p95_seconds": 0.20460932000059984,
+    "gpu_update_seconds": 0.9563332094874103,
+    "producer_batches_per_second": 11.565601911967418,
+    "gpu_batches_per_second": 1.0456606443019947,
+    "producer_headroom": 11.060569196125531,
     "sustainable": true,
-    "buffer_seconds": 60.03027336467184,
+    "buffer_seconds": 60.24899219770685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321109.4594113842,
+    "generation_bytes_per_second": 295712.7780192324,
     "upload_bytes_per_second": 4949702.305036583,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.90261494403977,
+    "estimated_full_shards_per_hour": 63.453078321769034,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -109,3 +109,4 @@
 {"step": 408000, "loss": 0.5376939177513123, "lr": 0.0003, "grad_norm": 0.16437156780921008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59720455009, "tokens_per_second": 151925.93550526028, "data_wait_s": 1.2930000593769364e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 408100, "loss": 0.5757578611373901, "lr": 0.0003, "grad_norm": 0.15850202302499988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59735165744, "tokens_per_second": 153632.57055071363, "data_wait_s": 8.314998922287486e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 408200, "loss": 0.5943154096603394, "lr": 0.0003, "grad_norm": 0.15907940341035445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59749757703, "tokens_per_second": 150970.2815007796, "data_wait_s": 1.5987001461326145e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 408300, "loss": 0.5720790028572083, "lr": 0.0003, "grad_norm": 0.15951854378604843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59764349486, "tokens_per_second": 151496.766827079, "data_wait_s": 8.944000001065433e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd843a9689890acc67ed7d4c0badc73d83ff2030f0dbe4c59be7c554f5e8be0c
-size 236829
+oid sha256:ca4064cbb55abfb79bb5fab5e6d5d45244f30f121a94e69b137ae3e4ba869542
+size 238947