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": 76878.52291007526,
-      "required_pages_per_second_upper_estimate": 0.005318634597668505
+      "required_token_positions_per_second": 76333.13007155815,
+      "required_pages_per_second_upper_estimate": 0.005280903055614156
     },
     {
       "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": 9609.815363759408,
-      "required_pages_per_second_upper_estimate": 0.004023429017041667
+      "required_token_positions_per_second": 9541.641258944768,
+      "required_pages_per_second_upper_estimate": 0.003994885942992973
     },
     {
       "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": 19219.630727518816,
-      "required_pages_per_second_upper_estimate": 0.0040029422247953115
+      "required_token_positions_per_second": 19083.282517889536,
+      "required_pages_per_second_upper_estimate": 0.003974544488473624
     },
     {
       "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": 28829.44609127822,
-      "required_pages_per_second_upper_estimate": 0.0017282812731567894
+      "required_token_positions_per_second": 28624.923776834305,
+      "required_pages_per_second_upper_estimate": 0.0017160204726933688
     },
     {
       "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": 28829.44609127822,
-      "required_pages_per_second_upper_estimate": 0.002241362626264938
+      "required_token_positions_per_second": 28624.923776834305,
+      "required_pages_per_second_upper_estimate": 0.0022254619159155125
     },
     {
       "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": 28829.44609127822,
-      "required_pages_per_second_upper_estimate": 0.36979317981142135
+      "required_token_positions_per_second": 28624.923776834305,
+      "required_pages_per_second_upper_estimate": 0.3671697871606868
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0844003743202677,
-    "producer_p95_seconds": 0.23067088199968566,
-    "gpu_update_seconds": 0.9590194661549309,
-    "producer_batches_per_second": 11.848288684187297,
-    "gpu_batches_per_second": 1.0427317017968107,
-    "producer_headroom": 11.36273948875881,
+    "producer_batch_seconds": 0.0890161605467199,
+    "producer_p95_seconds": 0.24352102299963008,
+    "gpu_update_seconds": 0.9658715675733986,
+    "producer_batches_per_second": 11.233915211105433,
+    "gpu_batches_per_second": 1.0353343379931388,
+    "producer_headroom": 10.850519294937051,
     "sustainable": true,
-    "buffer_seconds": 60.41822636776065,
+    "buffer_seconds": 60.84990875712411,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438026.7985897508,
-    "upload_bytes_per_second": 4866530.592281519,
+    "generation_bytes_per_second": 478093.24722499104,
+    "upload_bytes_per_second": 4913880.672516342,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.99035423535722,
+    "estimated_full_shards_per_hour": 102.58768141329097,
     "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": 0.9255875790004211,
-      "bytes": 259525,
-      "pages": 8,
-      "shards": 1
-    },
     {
       "seconds": 0.9178394819991809,
       "bytes": 643385,
@@ -562,6 +556,12 @@
       "bytes": 7694709,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2034739760019875,
+      "bytes": 5697150,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -206,3 +206,4 @@
 {"step": 417700, "loss": 0.6011258363723755, "lr": 0.0003, "grad_norm": 0.19596748567520328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61140454233, "tokens_per_second": 149981.84880119187, "data_wait_s": 1.3928998669143766e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 417800, "loss": 0.5787203311920166, "lr": 0.0003, "grad_norm": 0.1585995637576139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61154935523, "tokens_per_second": 150605.2872722367, "data_wait_s": 9.880001016426831e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 417900, "loss": 0.629663348197937, "lr": 0.0003, "grad_norm": 0.17760820649849354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61169583518, "tokens_per_second": 151079.328930287, "data_wait_s": 1.5313999028876424e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 418000, "loss": 0.5443013906478882, "lr": 0.0003, "grad_norm": 0.13174783431493042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61184254096, "tokens_per_second": 149585.14254806645, "data_wait_s": 1.0854000720428303e-05, "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:f7c758e8edcdc5e1de0d2a86680b2a24062283107d1b8fb81fced73889747220
-size 443841
+oid sha256:c682a5d58917400ecb5fd4bc6ef2591e27b9acdaf487958f2ae03c20e035ab23
+size 445959