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": 76200.4259541335,
-      "required_pages_per_second_upper_estimate": 0.0052717222768547435
+      "required_token_positions_per_second": 76770.96320747986,
+      "required_pages_per_second_upper_estimate": 0.0053111933678700604
     },
     {
       "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": 9525.053244266688,
-      "required_pages_per_second_upper_estimate": 0.003987940887644398
+      "required_token_positions_per_second": 9596.370400934982,
+      "required_pages_per_second_upper_estimate": 0.0040177998918698304
     },
     {
       "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": 19050.106488533376,
-      "required_pages_per_second_upper_estimate": 0.003967634796469465
+      "required_token_positions_per_second": 19192.740801869964,
+      "required_pages_per_second_upper_estimate": 0.0039973417624176825
     },
     {
       "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": 28575.159732800068,
-      "required_pages_per_second_upper_estimate": 0.001713037194238812
+      "required_token_positions_per_second": 28789.111202804947,
+      "required_pages_per_second_upper_estimate": 0.0017258632581806253
     },
     {
       "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": 28575.159732800068,
-      "required_pages_per_second_upper_estimate": 0.002221592980380747
+      "required_token_positions_per_second": 28789.111202804947,
+      "required_pages_per_second_upper_estimate": 0.0022382267661005656
     },
     {
       "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": 28575.159732800068,
-      "required_pages_per_second_upper_estimate": 0.3665314674362831
+      "required_token_positions_per_second": 28789.111202804947,
+      "required_pages_per_second_upper_estimate": 0.3692758071703152
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09069118620311656,
+    "producer_batch_seconds": 0.23548812513269013,
     "producer_p95_seconds": 0.2813108669979556,
-    "gpu_update_seconds": 0.967553646542321,
-    "producer_batches_per_second": 11.026429820427637,
-    "gpu_batches_per_second": 1.0335344232060208,
-    "producer_headroom": 10.668662381097752,
+    "gpu_update_seconds": 0.9603630971874614,
+    "producer_batches_per_second": 4.246498626784393,
+    "gpu_batches_per_second": 1.041272829962563,
+    "producer_headroom": 4.078180573420962,
     "sustainable": true,
-    "buffer_seconds": 60.955879732166224,
+    "buffer_seconds": 60.50287512281007,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 533918.4810342124,
+    "generation_bytes_per_second": 512071.77260747476,
     "upload_bytes_per_second": 4834118.412739204,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.56647704381732,
+    "estimated_full_shards_per_hour": 109.8786819807833,
     "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": [

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 464100, "loss": 0.5820420384407043, "lr": 0.0003, "grad_norm": 0.13034105932482143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67930203612, "tokens_per_second": 151317.34910678276, "data_wait_s": 1.1707998055499047e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 464200, "loss": 0.5661742687225342, "lr": 0.0003, "grad_norm": 0.16984531684949464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67944768634, "tokens_per_second": 150374.72092619466, "data_wait_s": 9.531999239698052e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 464300, "loss": 0.5670545101165771, "lr": 0.0003, "grad_norm": 0.1584016884536606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67959340525, "tokens_per_second": 151428.0759303254, "data_wait_s": 1.302200325881131e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 464400, "loss": 0.5809440612792969, "lr": 0.0003, "grad_norm": 0.17934050200999524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67973966158, "tokens_per_second": 150585.80848647223, "data_wait_s": 1.7591999494470656e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:899c7a464b98d75a68bd5c25b95541a8fe0308b62bbc498d92129d3fd3ebebc1
-size 513735
+oid sha256:9dec120111a5ce65a01907affa61a287030cbeb6c3016bde49cb3d3490d8b1e0
+size 515853