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": 77201.62098354206,
-      "required_pages_per_second_upper_estimate": 0.005340987272081756
+      "required_token_positions_per_second": 75886.36432581821,
+      "required_pages_per_second_upper_estimate": 0.005249994764684515
     },
     {
       "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": 9650.202622942757,
-      "required_pages_per_second_upper_estimate": 0.00404033831907986
+      "required_token_positions_per_second": 9485.795540727277,
+      "required_pages_per_second_upper_estimate": 0.003971504506966518
     },
     {
       "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": 19300.405245885515,
-      "required_pages_per_second_upper_estimate": 0.004019765426803799
+      "required_token_positions_per_second": 18971.591081454553,
+      "required_pages_per_second_upper_estimate": 0.003951282107765472
     },
     {
       "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": 28950.607868828272,
-      "required_pages_per_second_upper_estimate": 0.0017355447367175929
+      "required_token_positions_per_second": 28457.386622181828,
+      "required_pages_per_second_upper_estimate": 0.0017059768760863759
     },
     {
       "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": 28950.607868828272,
-      "required_pages_per_second_upper_estimate": 0.0022507824215351174
+      "required_token_positions_per_second": 28457.386622181828,
+      "required_pages_per_second_upper_estimate": 0.0022124366390593457
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28950.607868828272,
-      "required_pages_per_second_upper_estimate": 0.37134731300045243
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28457.386622181828,
+      "required_pages_per_second_upper_estimate": 0.36502080042818624
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11534706257805283,
-    "producer_p95_seconds": 0.356001396998181,
-    "gpu_update_seconds": 0.9550058542905132,
-    "producer_batches_per_second": 8.669488217988404,
-    "gpu_batches_per_second": 1.0471139998852819,
-    "producer_headroom": 8.279412001881553,
+    "producer_batch_seconds": 0.4194785052658858,
+    "producer_p95_seconds": 0.3837743109979783,
+    "gpu_update_seconds": 0.9715579426555307,
+    "producer_batches_per_second": 2.3839123755963407,
+    "gpu_batches_per_second": 1.0292746897490535,
+    "producer_headroom": 2.31610900310544,
     "sustainable": true,
-    "buffer_seconds": 60.165368820302334,
+    "buffer_seconds": 60.23659244464291,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341725.07458227786,
+    "generation_bytes_per_second": 317954.5595464571,
     "upload_bytes_per_second": 5131831.486766642,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.32624605275394,
+    "estimated_full_shards_per_hour": 68.2256468753365,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -408,3 +408,4 @@
 {"step": 437900, "loss": 0.575453519821167, "lr": 0.0003, "grad_norm": 0.18456594881201405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64095892710, "tokens_per_second": 151770.3124992193, "data_wait_s": 8.305003575515002e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 438000, "loss": 0.5736438035964966, "lr": 0.0003, "grad_norm": 0.14985151077977857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64110620537, "tokens_per_second": 152670.74583348134, "data_wait_s": 2.0515995856840163e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 438100, "loss": 0.556441605091095, "lr": 0.0003, "grad_norm": 0.169631654008228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64125325344, "tokens_per_second": 152286.5565314529, "data_wait_s": 8.811002771835774e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 438200, "loss": 0.5989417433738708, "lr": 0.0003, "grad_norm": 0.1770277063586694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64139964316, "tokens_per_second": 152261.89422793503, "data_wait_s": 1.0154995834454894e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cc97cab01588c063ec95b7666fc21562dcbc8ee6df667476dacc08dd00271331
-size 874809
+oid sha256:e5161b073bf162908e271e3f0430c309d86afae28ab8db68f422458eaf511521
+size 876927