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": 76063.66084404741,
-      "required_pages_per_second_upper_estimate": 0.005262260549200212
+      "required_token_positions_per_second": 76431.70005325026,
+      "required_pages_per_second_upper_estimate": 0.005287722355661484
     },
     {
       "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": 9507.957605505926,
-      "required_pages_per_second_upper_estimate": 0.0039807833006927995
+      "required_token_positions_per_second": 9553.962506656282,
+      "required_pages_per_second_upper_estimate": 0.004000044592112874
     },
     {
       "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": 19015.915211011852,
-      "required_pages_per_second_upper_estimate": 0.003960513655046356
+      "required_token_positions_per_second": 19107.925013312564,
+      "required_pages_per_second_upper_estimate": 0.003979676870403935
     },
     {
       "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": 28523.872816517774,
-      "required_pages_per_second_upper_estimate": 0.0017099626219182753
+      "required_token_positions_per_second": 28661.887519968845,
+      "required_pages_per_second_upper_estimate": 0.0017182363926539167
     },
     {
       "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": 28523.872816517774,
-      "required_pages_per_second_upper_estimate": 0.0022176056482271027
+      "required_token_positions_per_second": 28661.887519968845,
+      "required_pages_per_second_upper_estimate": 0.0022283356843578995
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28523.872816517774,
-      "required_pages_per_second_upper_estimate": 0.36587361394181417
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28661.887519968845,
+      "required_pages_per_second_upper_estimate": 0.3676439183690415
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.38222357247650507,
-    "producer_p95_seconds": 0.2986995319988637,
-    "gpu_update_seconds": 0.9692933416807773,
-    "producer_batches_per_second": 2.616269827422716,
-    "gpu_batches_per_second": 1.0316794276807644,
-    "producer_headroom": 2.535932923761155,
+    "producer_batch_seconds": 0.10626844970320803,
+    "producer_p95_seconds": 0.2982578849987476,
+    "gpu_update_seconds": 0.9646259333317645,
+    "producer_batches_per_second": 9.410130690650435,
+    "gpu_batches_per_second": 1.0366712789340584,
+    "producer_headroom": 9.077256100242558,
     "sustainable": true,
-    "buffer_seconds": 60.09618718420819,
+    "buffer_seconds": 60.77143379990116,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334373.38381038327,
-    "upload_bytes_per_second": 5141106.712311341,
+    "generation_bytes_per_second": 347661.94227066403,
+    "upload_bytes_per_second": 5080571.87659806,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.74874435170769,
+    "estimated_full_shards_per_hour": 74.60015965547505,
     "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": [
-    {
-      "seconds": 1.3833632300011232,
-      "bytes": 11874470,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.439283288000297,
       "bytes": 6018491,
@@ -562,6 +556,12 @@
       "bytes": 5834868,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.0089386899999226,
+      "bytes": 4885086,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -272,3 +272,4 @@
 {"step": 952200, "loss": 0.5294348001480103, "lr": 3e-05, "grad_norm": 0.3469594693218845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139376085306, "tokens_per_second": 153018.2428823329, "data_wait_s": 1.1668998922687024e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952300, "loss": 0.5251018404960632, "lr": 3e-05, "grad_norm": 0.34399387024216427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139390817641, "tokens_per_second": 152716.52592594008, "data_wait_s": 1.2970998795935884e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 952400, "loss": 0.5081045627593994, "lr": 3e-05, "grad_norm": 0.3509164334995831, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139405519742, "tokens_per_second": 152630.40931622678, "data_wait_s": 1.0414998541818932e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 952500, "loss": 0.5086105465888977, "lr": 3e-05, "grad_norm": 0.3440548034632097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139420057862, "tokens_per_second": 150938.92247970775, "data_wait_s": 1.4551002095686272e-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 sha256:4263dcc4a2a7142b98f75977a283b44921b525716d23e2190fd7e459ddd57e1c
-size 584412
+oid sha256:95907461e3aa0cdafe09943e0252c987af1473c4d46e0c47488b42fd9dd0560e
+size 586530