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": 75687.10192082937,
-      "required_pages_per_second_upper_estimate": 0.0052362093291549085
+      "required_token_positions_per_second": 76110.80031888915,
+      "required_pages_per_second_upper_estimate": 0.005265521767448414
     },
     {
       "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": 9460.887740103672,
-      "required_pages_per_second_upper_estimate": 0.003961076130979437
+      "required_token_positions_per_second": 9513.850039861143,
+      "required_pages_per_second_upper_estimate": 0.003983250339909307
     },
     {
       "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": 18921.775480207343,
-      "required_pages_per_second_upper_estimate": 0.003940906831751427
+      "required_token_positions_per_second": 19027.700079722286,
+      "required_pages_per_second_upper_estimate": 0.003962968132410852
     },
     {
       "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": 28382.663220311013,
-      "required_pages_per_second_upper_estimate": 0.0017014973222402488
+      "required_token_positions_per_second": 28541.55011958343,
+      "required_pages_per_second_upper_estimate": 0.0017110223492453824
     },
     {
       "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": 28382.663220311013,
-      "required_pages_per_second_upper_estimate": 0.0022066272232373996
+      "required_token_positions_per_second": 28541.55011958343,
+      "required_pages_per_second_upper_estimate": 0.0022189799807862186
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28382.663220311013,
-      "required_pages_per_second_upper_estimate": 0.3640623288607254
+      "required_token_positions_per_second": 28541.55011958343,
+      "required_pages_per_second_upper_estimate": 0.36610035940513114
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.45224705421085787,
-    "producer_p95_seconds": 0.23327610200067284,
-    "gpu_update_seconds": 0.9741157757251871,
-    "producer_batches_per_second": 2.211180793084293,
-    "gpu_batches_per_second": 1.0265720204105546,
-    "producer_headroom": 2.1539460935239405,
+    "producer_batch_seconds": 0.09076385435159295,
+    "producer_p95_seconds": 0.22895767299996805,
+    "gpu_update_seconds": 0.9686930066573248,
+    "producer_batches_per_second": 11.017601744041068,
+    "gpu_batches_per_second": 1.0323187977279886,
+    "producer_headroom": 10.672673759588127,
     "sustainable": true,
-    "buffer_seconds": 60.3951780949616,
+    "buffer_seconds": 60.05896641275414,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268144.05876303563,
+    "generation_bytes_per_second": 268899.7435005941,
     "upload_bytes_per_second": 4805712.845759498,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.53747293632795,
+    "estimated_full_shards_per_hour": 57.699625289567635,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 944900, "loss": 0.5289706587791443, "lr": 3e-05, "grad_norm": 0.35097219719823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138306891113, "tokens_per_second": 152803.21131051768, "data_wait_s": 1.2256998161319643e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 945000, "loss": 0.547132670879364, "lr": 3e-05, "grad_norm": 0.3489687738519755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138321369205, "tokens_per_second": 150205.0161449377, "data_wait_s": 1.516900010756217e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 945100, "loss": 0.5197945237159729, "lr": 3e-05, "grad_norm": 0.34259761542248635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138336025394, "tokens_per_second": 141410.8378124374, "data_wait_s": 1.3495999155566096e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 945200, "loss": 0.5494606494903564, "lr": 3e-05, "grad_norm": 0.3465691098478906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138350543162, "tokens_per_second": 151311.5287856669, "data_wait_s": 1.5022000297904015e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:95fd5255d4802f1009a9c97ea45eedfda4b4d52ff2372e5a59689b44672c7c86
-size 429015
+oid sha256:972e2e92dad5053da2b493a344a04ad0bb2ad9501f64c73aa2030ce7d99c765a
+size 431133