Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75876.88957548271,
-      "required_pages_per_second_upper_estimate": 0.006576816480184379
+      "required_token_positions_per_second": 75797.65445524092,
+      "required_pages_per_second_upper_estimate": 0.006569948580781397
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9484.611196935339,
-      "required_pages_per_second_upper_estimate": 0.0037835532140319686
+      "required_token_positions_per_second": 9474.706806905115,
+      "required_pages_per_second_upper_estimate": 0.003779602204765085
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18969.222393870677,
-      "required_pages_per_second_upper_estimate": 0.004388195753910137
+      "required_token_positions_per_second": 18949.41361381023,
+      "required_pages_per_second_upper_estimate": 0.004383613341265774
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28453.833590806014,
-      "required_pages_per_second_upper_estimate": 0.0017326891799526443
+      "required_token_positions_per_second": 28424.120420715342,
+      "required_pages_per_second_upper_estimate": 0.0017308798037870777
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28453.833590806014,
-      "required_pages_per_second_upper_estimate": 0.002190454875072827
+      "required_token_positions_per_second": 28424.120420715342,
+      "required_pages_per_second_upper_estimate": 0.002188167472987928
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28453.833590806014,
-      "required_pages_per_second_upper_estimate": 0.353301384342675
+      "required_token_positions_per_second": 28424.120420715342,
+      "required_pages_per_second_upper_estimate": 0.3529324462137321
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13512281224234357,
-    "producer_p95_seconds": 0.35335284499888076,
-    "gpu_update_seconds": 0.9716792611359618,
-    "producer_batches_per_second": 7.400674863149637,
-    "gpu_batches_per_second": 1.0291461802230184,
-    "producer_headroom": 7.191082282932724,
+    "producer_batch_seconds": 0.08934438057812599,
+    "producer_p95_seconds": 0.3562607640014903,
+    "gpu_update_seconds": 0.9726950065920171,
+    "producer_batches_per_second": 11.192645732493085,
+    "gpu_batches_per_second": 1.028071485124253,
+    "producer_headroom": 10.887030614549474,
     "sustainable": true,
-    "buffer_seconds": 60.244114190429634,
+    "buffer_seconds": 60.307090408705065,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326895.4574641931,
-    "upload_bytes_per_second": 6252487.61413573,
+    "generation_bytes_per_second": 327197.997289838,
+    "upload_bytes_per_second": 6195647.0997435795,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.14415543503137,
+    "estimated_full_shards_per_hour": 70.20907343884807,
     "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": 1.2109974509985477,
-      "bytes": 11754963,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.878410489000089,
       "bytes": 971318,
@@ -562,6 +556,12 @@
       "bytes": 7008906,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 0.9077771720003511,
+      "bytes": 6012024,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -294,3 +294,4 @@
 {"step": 29400, "loss": 0.667667131870985, "lr": 0.0003, "grad_norm": 0.12892447224017894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4302056247, "tokens_per_second": 149231.15741604165, "data_wait_s": 1.2896001862827688e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 29500, "loss": 0.6745256744325161, "lr": 0.0003, "grad_norm": 0.13999687860383306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4316666336, "tokens_per_second": 149070.9060663636, "data_wait_s": 9.485000191489235e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 29600, "loss": 0.6538167893886566, "lr": 0.0003, "grad_norm": 0.14755425168421338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4331378206, "tokens_per_second": 150294.98719142217, "data_wait_s": 1.3083001249469817e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 29700, "loss": 0.7051474638283253, "lr": 0.0003, "grad_norm": 0.14295477642627422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4346017764, "tokens_per_second": 148882.31375901506, "data_wait_s": 1.1593001545406878e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:52a0357694ea37ba7ab042ad5c444bb56d750808f814f2262061693324b99b59
-size 625234
+oid sha256:0b6d9902350052a533ee6fa410d0d1c63e84873469fe7ae17d88eae978e3b891
+size 627352