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": 76455.55900804087,
-      "required_pages_per_second_upper_estimate": 0.005289372973514301
+      "required_token_positions_per_second": 77340.83086633166,
+      "required_pages_per_second_upper_estimate": 0.00535061813478457
     },
     {
       "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": 9556.94487600511,
-      "required_pages_per_second_upper_estimate": 0.004001293247880273
+      "required_token_positions_per_second": 9667.603858291457,
+      "required_pages_per_second_upper_estimate": 0.004047623852941245
     },
     {
       "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": 19113.88975201022,
-      "required_pages_per_second_upper_estimate": 0.003980919168173919
+      "required_token_positions_per_second": 19335.207716582914,
+      "required_pages_per_second_upper_estimate": 0.004027013863647195
     },
     {
       "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": 28670.834628015327,
-      "required_pages_per_second_upper_estimate": 0.0017187727581198604
+      "required_token_positions_per_second": 29002.81157487437,
+      "required_pages_per_second_upper_estimate": 0.0017386742691846127
     },
     {
       "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": 28670.834628015327,
-      "required_pages_per_second_upper_estimate": 0.002229031282654344
+      "required_token_positions_per_second": 29002.81157487437,
+      "required_pages_per_second_upper_estimate": 0.0022548410300603657
     },
     {
       "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": 28670.834628015327,
-      "required_pages_per_second_upper_estimate": 0.3677586822644056
+      "required_token_positions_per_second": 29002.81157487437,
+      "required_pages_per_second_upper_estimate": 0.3720169260896393
     }
   ],
   "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": 0.10894605182801342,
-    "producer_p95_seconds": 0.34728073600126663,
-    "gpu_update_seconds": 0.9643249092227026,
-    "producer_batches_per_second": 9.178854884788665,
-    "gpu_batches_per_second": 1.0369948867193044,
-    "producer_headroom": 8.85139840354219,
+    "producer_batch_seconds": 0.10972448413264146,
+    "producer_p95_seconds": 0.3456271570030367,
+    "gpu_update_seconds": 0.9532868883633314,
+    "producer_batches_per_second": 9.113736172057466,
+    "gpu_batches_per_second": 1.049002154762528,
+    "producer_headroom": 8.688005196825001,
     "sustainable": true,
-    "buffer_seconds": 60.752469281030265,
+    "buffer_seconds": 60.05707396688988,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396931.5223456809,
+    "generation_bytes_per_second": 413214.8442076041,
     "upload_bytes_per_second": 4786988.736817365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.17226460244962,
+    "estimated_full_shards_per_hour": 88.66628641768544,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -363,3 +363,4 @@
 {"step": 961300, "loss": 0.5140148997306824, "lr": 3e-05, "grad_norm": 0.35353837122117165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140708527642, "tokens_per_second": 151693.61127884014, "data_wait_s": 1.3297998521011323e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 961400, "loss": 0.4920903146266937, "lr": 3e-05, "grad_norm": 0.3526757492499405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140723051873, "tokens_per_second": 150683.633431203, "data_wait_s": 1.4518998796120286e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 961500, "loss": 0.5077930688858032, "lr": 3e-05, "grad_norm": 0.3514096955737956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140737578155, "tokens_per_second": 150925.29597307602, "data_wait_s": 1.22560013551265e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 961600, "loss": 0.48917630314826965, "lr": 3e-05, "grad_norm": 0.34890945974183674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140752217671, "tokens_per_second": 152169.75341136171, "data_wait_s": 8.320996130350977e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0cac8fb9fc95791e58dccaea1b7d4bacefa8fab082bf99dc3767968ac81c3985
-size 778716
+oid sha256:a5d3491dbce3002b60900cf436e56e6bfd66da50854910ce71d196556f6bb108
+size 780834