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": 77340.83086633166,
-      "required_pages_per_second_upper_estimate": 0.00535061813478457
+      "required_token_positions_per_second": 77538.17046717224,
+      "required_pages_per_second_upper_estimate": 0.005364270546261681
     },
     {
       "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": 9667.603858291457,
-      "required_pages_per_second_upper_estimate": 0.004047623852941245
+      "required_token_positions_per_second": 9692.27130839653,
+      "required_pages_per_second_upper_estimate": 0.004057951599185305
     },
     {
       "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": 19335.207716582914,
-      "required_pages_per_second_upper_estimate": 0.004027013863647195
+      "required_token_positions_per_second": 19384.54261679306,
+      "required_pages_per_second_upper_estimate": 0.004037289022312158
     },
     {
       "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": 29002.81157487437,
-      "required_pages_per_second_upper_estimate": 0.0017386742691846127
+      "required_token_positions_per_second": 29076.81392518959,
+      "required_pages_per_second_upper_estimate": 0.0017431105970909637
     },
     {
       "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": 29002.81157487437,
-      "required_pages_per_second_upper_estimate": 0.0022548410300603657
+      "required_token_positions_per_second": 29076.81392518959,
+      "required_pages_per_second_upper_estimate": 0.0022605943872954348
     },
     {
       "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": 29002.81157487437,
-      "required_pages_per_second_upper_estimate": 0.3720169260896393
+      "required_token_positions_per_second": 29076.81392518959,
+      "required_pages_per_second_upper_estimate": 0.3729661487819498
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10972448413264146,
+    "producer_batch_seconds": 0.4122696186486223,
     "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,
+    "gpu_update_seconds": 0.9508607122889833,
+    "producer_batches_per_second": 2.4255971208305325,
+    "gpu_batches_per_second": 1.051678744400665,
+    "producer_headroom": 2.3064050060390273,
     "sustainable": true,
-    "buffer_seconds": 60.05707396688988,
+    "buffer_seconds": 60.85508558649493,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 413214.8442076041,
-    "upload_bytes_per_second": 4786988.736817365,
+    "generation_bytes_per_second": 394859.2061229144,
+    "upload_bytes_per_second": 4767746.344361747,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.66628641768544,
+    "estimated_full_shards_per_hour": 84.72759378209662,
     "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": [
-    {
-      "seconds": 1.1317712890013354,
-      "bytes": 6287146,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 0.9238542080020125,
       "bytes": 660547,
@@ -562,6 +556,12 @@
       "bytes": 5901438,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.9387289110018173,
+      "bytes": 8423867,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -364,3 +364,4 @@
 {"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}
+{"step": 961700, "loss": 0.5648232102394104, "lr": 3e-05, "grad_norm": 0.35461705620083384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140766952259, "tokens_per_second": 138410.45664885247, "data_wait_s": 1.4290002582129091e-05, "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:a5d3491dbce3002b60900cf436e56e6bfd66da50854910ce71d196556f6bb108
-size 780834
+oid sha256:e44983335e818306eb5b442f571ec09b1b1b20c7d827726fd7d330b183877262
+size 782952