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": 77235.94945110014,
-      "required_pages_per_second_upper_estimate": 0.005343362195120442
+      "required_token_positions_per_second": 77119.6283293756,
+      "required_pages_per_second_upper_estimate": 0.005335314830017351
     },
     {
       "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": 9654.493681387517,
-      "required_pages_per_second_upper_estimate": 0.004042134895643191
+      "required_token_positions_per_second": 9639.95354117195,
+      "required_pages_per_second_upper_estimate": 0.004036047242567588
     },
     {
       "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": 19308.987362775035,
-      "required_pages_per_second_upper_estimate": 0.004021552855426243
+      "required_token_positions_per_second": 19279.9070823439,
+      "required_pages_per_second_upper_estimate": 0.004015496199911005
     },
     {
       "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": 28963.481044162552,
-      "required_pages_per_second_upper_estimate": 0.0017363164639226823
+      "required_token_positions_per_second": 28919.86062351585,
+      "required_pages_per_second_upper_estimate": 0.0017337014863094383
     },
     {
       "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": 28963.481044162552,
-      "required_pages_per_second_upper_estimate": 0.0022517832542942435
+      "required_token_positions_per_second": 28919.86062351585,
+      "required_pages_per_second_upper_estimate": 0.002248391958454916
     },
     {
       "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": 28963.481044162552,
-      "required_pages_per_second_upper_estimate": 0.3715124362715018
+      "required_token_positions_per_second": 28919.86062351585,
+      "required_pages_per_second_upper_estimate": 0.3709529203514046
     }
   ],
   "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": 0.11561486713293334,
-    "producer_p95_seconds": 0.31231984800251666,
-    "gpu_update_seconds": 0.9545813901941983,
-    "producer_batches_per_second": 8.649406644650687,
-    "gpu_batches_per_second": 1.0475796095255554,
-    "producer_headroom": 8.256562619205589,
+    "producer_batch_seconds": 0.10560610817177007,
+    "producer_p95_seconds": 0.29577020199940307,
+    "gpu_update_seconds": 0.956021204940329,
+    "producer_batches_per_second": 9.469149250093409,
+    "gpu_batches_per_second": 1.0460019033389703,
+    "producer_headroom": 9.052707475834113,
     "sustainable": true,
-    "buffer_seconds": 60.13862758223449,
+    "buffer_seconds": 60.22933591124073,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 227333.44084289653,
+    "generation_bytes_per_second": 252999.07967791654,
     "upload_bytes_per_second": 4681275.070577774,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.780464353229256,
+    "estimated_full_shards_per_hour": 54.28771298173067,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -307,3 +307,4 @@
 {"step": 873200, "loss": 0.49504029750823975, "lr": 3e-05, "grad_norm": 0.4155996004658534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127812119291, "tokens_per_second": 151725.32101300327, "data_wait_s": 1.5221001376630738e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 873300, "loss": 0.5646331906318665, "lr": 3e-05, "grad_norm": 0.32176437313481454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127826608135, "tokens_per_second": 150051.26434386065, "data_wait_s": 1.41079981403891e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 873400, "loss": 0.5732330083847046, "lr": 3e-05, "grad_norm": 0.319052623109754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127841244868, "tokens_per_second": 151732.2769361672, "data_wait_s": 1.0592000762699172e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 873500, "loss": 0.5143589973449707, "lr": 3e-05, "grad_norm": 0.29719932330998033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127855896756, "tokens_per_second": 151729.69983282857, "data_wait_s": 1.980800152523443e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e8c92a7c2c9ea4c0e8769fd052bf50381cfdd6cfcd05943312af8919754bad55
-size 659325
+oid sha256:87ac245d7109fa4d03823eb6ffe23fbef245a80e11238ffcb575de5acf75d210
+size 661443