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": 75107.87516395623,
-      "required_pages_per_second_upper_estimate": 0.00651016026004247
+      "required_token_positions_per_second": 75126.20852568044,
+      "required_pages_per_second_upper_estimate": 0.006511749349371245
     },
     {
       "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": 9388.484395494528,
-      "required_pages_per_second_upper_estimate": 0.0037452067957134704
+      "required_token_positions_per_second": 9390.776065710055,
+      "required_pages_per_second_upper_estimate": 0.0037461209772259676
     },
     {
       "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": 18776.968790989056,
-      "required_pages_per_second_upper_estimate": 0.004343721266431322
+      "required_token_positions_per_second": 18781.55213142011,
+      "required_pages_per_second_upper_estimate": 0.0043447815415760095
     },
     {
       "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": 28165.453186483584,
-      "required_pages_per_second_upper_estimate": 0.0017151283263444578
+      "required_token_positions_per_second": 28172.328197130166,
+      "required_pages_per_second_upper_estimate": 0.0017155469784224415
     },
     {
       "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": 28165.453186483584,
-      "required_pages_per_second_upper_estimate": 0.0021682545532600342
+      "required_token_positions_per_second": 28172.328197130166,
+      "required_pages_per_second_upper_estimate": 0.0021687838106108554
     },
     {
       "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": 28165.453186483584,
-      "required_pages_per_second_upper_estimate": 0.3497206648681176
+      "required_token_positions_per_second": 28172.328197130166,
+      "required_pages_per_second_upper_estimate": 0.3498060294911676
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.16158711314869834,
-    "producer_p95_seconds": 0.45463397900311975,
-    "gpu_update_seconds": 0.9816280894520842,
-    "producer_batches_per_second": 6.188612325042057,
-    "gpu_batches_per_second": 1.0187157547194583,
-    "producer_headroom": 6.074915692990656,
+    "producer_batch_seconds": 0.10853620750765458,
+    "producer_p95_seconds": 0.42890520799846854,
+    "gpu_update_seconds": 0.9813885386588824,
+    "producer_batches_per_second": 9.213515221908546,
+    "gpu_batches_per_second": 1.0189644168522194,
+    "producer_headroom": 9.042038239540195,
     "sustainable": true,
-    "buffer_seconds": 60.860941546029224,
+    "buffer_seconds": 60.84608939685071,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 258236.02319104638,
+    "generation_bytes_per_second": 269967.7498793661,
     "upload_bytes_per_second": 5527129.434853146,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.41143914984268,
+    "estimated_full_shards_per_hour": 57.92879459653604,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 356600, "loss": 0.6038533113896847, "lr": 0.0003, "grad_norm": 0.15770938922945837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52196046402, "tokens_per_second": 144129.5103082584, "data_wait_s": 1.1427000572439283e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356700, "loss": 0.6333474293351173, "lr": 0.0003, "grad_norm": 0.18712339975237002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52210646056, "tokens_per_second": 146762.10956158003, "data_wait_s": 1.3723998563364148e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356800, "loss": 0.5970645230263472, "lr": 0.0003, "grad_norm": 0.17870772044610614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52225208874, "tokens_per_second": 146988.22783714926, "data_wait_s": 1.4484001439996064e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 356900, "loss": 0.615576034411788, "lr": 0.0003, "grad_norm": 0.15941947816925933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52239903919, "tokens_per_second": 140987.3160980433, "data_wait_s": 1.2231997970957309e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8931f5455509a1055ff2eeb0662437f39f5936d1a6767e1875491fcadf4926cd
-size 891753
+oid sha256:24302e6da5ebe148e20c91b610b13c460d0e5fa2c047fb6cfd97a0eb8f85ee03
+size 893871