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": 76920.53103505312,
-      "required_pages_per_second_upper_estimate": 0.005321540817226752
+      "required_token_positions_per_second": 76433.94967302271,
+      "required_pages_per_second_upper_estimate": 0.00528787798957718
     },
     {
       "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": 9615.06637938164,
-      "required_pages_per_second_upper_estimate": 0.00402562750762902
+      "required_token_positions_per_second": 9554.243709127839,
+      "required_pages_per_second_upper_estimate": 0.0040001623257155386
     },
     {
       "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": 19230.13275876328,
-      "required_pages_per_second_upper_estimate": 0.004005129520946346
+      "required_token_positions_per_second": 19108.487418255678,
+      "required_pages_per_second_upper_estimate": 0.00397979400452197
     },
     {
       "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": 28845.19913814492,
-      "required_pages_per_second_upper_estimate": 0.0017292256442629382
+      "required_token_positions_per_second": 28662.731127383515,
+      "required_pages_per_second_upper_estimate": 0.0017182869656303123
     },
     {
       "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": 28845.19913814492,
-      "required_pages_per_second_upper_estimate": 0.0022425873563684936
+      "required_token_positions_per_second": 28662.731127383515,
+      "required_pages_per_second_upper_estimate": 0.0022284012711237533
     },
     {
       "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": 28845.19913814492,
-      "required_pages_per_second_upper_estimate": 0.36999524298232345
+      "required_token_positions_per_second": 28662.731127383515,
+      "required_pages_per_second_upper_estimate": 0.3676547392591618
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.2914910166247182,
-    "producer_p95_seconds": 0.2586335230007535,
-    "gpu_update_seconds": 0.9584957228961632,
-    "producer_batches_per_second": 3.4306374569596283,
-    "gpu_batches_per_second": 1.0433014734572092,
-    "producer_headroom": 3.288251329303174,
+    "producer_batch_seconds": 0.43217783876565363,
+    "producer_p95_seconds": 0.25035172199932276,
+    "gpu_update_seconds": 0.9645975422623257,
+    "producer_batches_per_second": 2.313862281453643,
+    "gpu_batches_per_second": 1.0367017913550172,
+    "producer_headroom": 2.231945869823682,
     "sustainable": true,
-    "buffer_seconds": 60.38523054245829,
+    "buffer_seconds": 60.76964516252652,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356342.5158592566,
+    "generation_bytes_per_second": 320589.2848913059,
     "upload_bytes_per_second": 5072144.418155734,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.46280867417596,
+    "estimated_full_shards_per_hour": 68.79099760107405,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -183,3 +183,4 @@
 {"step": 778300, "loss": 0.4791816771030426, "lr": 3e-05, "grad_norm": 0.24315523523943586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113921334291, "tokens_per_second": 153813.61661321798, "data_wait_s": 1.4381999790202826e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 778400, "loss": 0.5778656005859375, "lr": 3e-05, "grad_norm": 0.2760404048411358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113935975517, "tokens_per_second": 151290.53802068773, "data_wait_s": 1.3567998394137248e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 778500, "loss": 0.5702077150344849, "lr": 3e-05, "grad_norm": 0.25043204536375985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113950515945, "tokens_per_second": 137848.96088460786, "data_wait_s": 1.4463999832514673e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 778600, "loss": 0.537959098815918, "lr": 3e-05, "grad_norm": 0.25468201989668166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113965021281, "tokens_per_second": 150455.7726355637, "data_wait_s": 1.4880999515298754e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d31a55929d0446eb1e931b2c32939774de0eaf17dd9043aea7f2c5449cae492b
-size 394344
+oid sha256:5482d07fbe39e4ad1e4a63d72d6c1ca631d55ba0cc440a9590a52a602c67034d
+size 396462