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": 75184.06414226176,
-      "required_pages_per_second_upper_estimate": 0.0065167641275827946
+      "required_token_positions_per_second": 75728.85860641896,
+      "required_pages_per_second_upper_estimate": 0.006563985530966995
     },
     {
       "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": 9398.00801778272,
-      "required_pages_per_second_upper_estimate": 0.0037490059110350726
+      "required_token_positions_per_second": 9466.10732580237,
+      "required_pages_per_second_upper_estimate": 0.0037761717431794998
     },
     {
       "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": 18796.01603556544,
-      "required_pages_per_second_upper_estimate": 0.0043481275112440135
+      "required_token_positions_per_second": 18932.21465160474,
+      "required_pages_per_second_upper_estimate": 0.004379634664050931
     },
     {
       "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": 28194.024053348163,
-      "required_pages_per_second_upper_estimate": 0.001716868142236757
+      "required_token_positions_per_second": 28398.32197740711,
+      "required_pages_per_second_upper_estimate": 0.0017293088139435778
     },
     {
       "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": 28194.024053348163,
-      "required_pages_per_second_upper_estimate": 0.002170454017680491
+      "required_token_positions_per_second": 28398.32197740711,
+      "required_pages_per_second_upper_estimate": 0.002186181437407398
     },
     {
       "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": 28194.024053348163,
-      "required_pages_per_second_upper_estimate": 0.3500754194140353
+      "required_token_positions_per_second": 28398.32197740711,
+      "required_pages_per_second_upper_estimate": 0.35261211588967944
     }
   ],
   "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": 0.4519897058827951,
-    "producer_p95_seconds": 0.3545515229998273,
-    "gpu_update_seconds": 0.98063334086986,
-    "producer_batches_per_second": 2.212439767951947,
-    "gpu_batches_per_second": 1.0197491338739932,
-    "producer_headroom": 2.169592201120056,
+    "producer_batch_seconds": 0.0855757972734068,
+    "producer_p95_seconds": 0.3539902830016217,
+    "gpu_update_seconds": 0.9735786509497271,
+    "producer_batches_per_second": 11.685546987135766,
+    "gpu_batches_per_second": 1.0271383817059863,
+    "producer_headroom": 11.376799071345287,
     "sustainable": true,
-    "buffer_seconds": 60.799267133931316,
+    "buffer_seconds": 60.36187635888308,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338971.3244651703,
-    "upload_bytes_per_second": 6118542.312796393,
+    "generation_bytes_per_second": 337847.08767189557,
+    "upload_bytes_per_second": 6237899.789747024,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.73535538164455,
+    "estimated_full_shards_per_hour": 72.49412033670092,
     "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": 0.8297740179987159,
-      "bytes": 928094,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 0.8634128480007348,
       "bytes": 6113242,
@@ -562,6 +556,12 @@
       "bytes": 2571680,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0315104050023365,
+      "bytes": 10269533,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -288,3 +288,4 @@
 {"step": 28800, "loss": 0.6906627379357815, "lr": 0.0003, "grad_norm": 0.15367915047919362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4214172372, "tokens_per_second": 148736.3276621741, "data_wait_s": 1.518800127087161e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 28900, "loss": 0.6480767019093037, "lr": 0.0003, "grad_norm": 0.14121009485425134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4228921148, "tokens_per_second": 150899.2919792501, "data_wait_s": 1.1834999895654619e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 29000, "loss": 0.6972436346113682, "lr": 0.0003, "grad_norm": 0.14391108421625456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4243645074, "tokens_per_second": 149921.3201087006, "data_wait_s": 1.4084002032177523e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 29100, "loss": 0.6293734144419432, "lr": 0.0003, "grad_norm": 0.13333394917548458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4258227975, "tokens_per_second": 149456.71999855785, "data_wait_s": 1.162899934570305e-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:af1d1f476bf4c83981b8e9c14b4a4f35909a7124e92c2cbb67d8d9b5e45818d4
-size 612526
+oid sha256:2ef7da1d8fa95654a6e90b6617302aa3060afadd29ee02610e6e7fe21f5f0c0b
+size 614644