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": 76895.3571815173,
-      "required_pages_per_second_upper_estimate": 0.005319799231627757
+      "required_token_positions_per_second": 77233.79034655003,
+      "required_pages_per_second_upper_estimate": 0.0053432128232578525
     },
     {
       "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": 9611.919647689663,
-      "required_pages_per_second_upper_estimate": 0.004024310036780819
+      "required_token_positions_per_second": 9654.223793318753,
+      "required_pages_per_second_upper_estimate": 0.004042021899144703
     },
     {
       "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": 19223.839295379326,
-      "required_pages_per_second_upper_estimate": 0.004003818758493258
+      "required_token_positions_per_second": 19308.447586637507,
+      "required_pages_per_second_upper_estimate": 0.004021440434291651
     },
     {
       "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": 28835.75894306899,
-      "required_pages_per_second_upper_estimate": 0.0017286597189824797
+      "required_token_positions_per_second": 28962.671379956264,
+      "required_pages_per_second_upper_estimate": 0.001736267925789801
     },
     {
       "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": 28835.75894306899,
-      "required_pages_per_second_upper_estimate": 0.00224185342272437
+      "required_token_positions_per_second": 28962.671379956264,
+      "required_pages_per_second_upper_estimate": 0.002251720306463541
     },
     {
       "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": 28835.75894306899,
-      "required_pages_per_second_upper_estimate": 0.3698741542959812
+      "required_token_positions_per_second": 28962.671379956264,
+      "required_pages_per_second_upper_estimate": 0.37150205076841325
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12103460697653645,
-    "producer_p95_seconds": 0.35878493200289086,
-    "gpu_update_seconds": 0.9588095133749035,
-    "producer_batches_per_second": 8.26209978269982,
-    "gpu_batches_per_second": 1.0429600312163263,
-    "producer_headroom": 7.9217798721053105,
+    "producer_batch_seconds": 0.13133296921870397,
+    "producer_p95_seconds": 0.37477072100591613,
+    "gpu_update_seconds": 0.9546080759364591,
+    "producer_batches_per_second": 7.614234308026165,
+    "gpu_batches_per_second": 1.0475503247958717,
+    "producer_headroom": 7.268609562514234,
     "sustainable": true,
-    "buffer_seconds": 60.40499934261892,
+    "buffer_seconds": 60.14030878399693,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308815.483631115,
-    "upload_bytes_per_second": 4955804.746962364,
+    "generation_bytes_per_second": 322247.5395725463,
+    "upload_bytes_per_second": 5011841.984188898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.26461393070305,
+    "estimated_full_shards_per_hour": 69.14682045347493,
     "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": 1.2987593140060198,
-      "bytes": 2393188,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.808516741002677,
       "bytes": 10401411,
@@ -562,6 +556,12 @@
       "bytes": 5442288,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0000966459992924,
+      "bytes": 5817685,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -419,3 +419,4 @@
 {"step": 686900, "loss": 0.5643567442893982, "lr": 0.0001, "grad_norm": 0.20095767760768526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100544126019, "tokens_per_second": 140151.1358112805, "data_wait_s": 1.2346994481049478e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687000, "loss": 0.5700095295906067, "lr": 0.0001, "grad_norm": 0.2111146027716485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100558815263, "tokens_per_second": 152244.84273307477, "data_wait_s": 1.2853997759521008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687100, "loss": 0.5151077508926392, "lr": 0.0001, "grad_norm": 0.20409389195283212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100573505138, "tokens_per_second": 151924.68522414909, "data_wait_s": 1.2672004231717438e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 687200, "loss": 0.587519645690918, "lr": 0.0001, "grad_norm": 0.21587036717376892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100588032378, "tokens_per_second": 150816.02977310724, "data_wait_s": 1.3817996659781784e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f29a357744076b93f6761f89d464c7108164fbd7dba06f5cdd061f29b084f9dd
-size 898107
+oid sha256:af8a2c644da4f1105ed5710fa0f0bab7ed337534f98aa55525fedd4dce798462
+size 900225