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": 75892.38885117188,
-      "required_pages_per_second_upper_estimate": 0.0052504115553273075
+      "required_token_positions_per_second": 77372.04882257793,
+      "required_pages_per_second_upper_estimate": 0.005352777865433325
     },
     {
       "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": 9486.548606396485,
-      "required_pages_per_second_upper_estimate": 0.0039718197998364155
+      "required_token_positions_per_second": 9671.50610282224,
+      "required_pages_per_second_upper_estimate": 0.004049257641238152
     },
     {
       "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": 18973.09721279297,
-      "required_pages_per_second_upper_estimate": 0.00395159579520391
+      "required_token_positions_per_second": 19343.01220564448,
+      "required_pages_per_second_upper_estimate": 0.004028639332900502
     },
     {
       "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": 28459.645819189456,
-      "required_pages_per_second_upper_estimate": 0.0017061123114973892
+      "required_token_positions_per_second": 29014.518308466722,
+      "required_pages_per_second_upper_estimate": 0.001739376069988327
     },
     {
       "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": 28459.645819189456,
-      "required_pages_per_second_upper_estimate": 0.002212612281689531
+      "required_token_positions_per_second": 29014.518308466722,
+      "required_pages_per_second_upper_estimate": 0.002255751177104692
     },
     {
       "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": 28459.645819189456,
-      "required_pages_per_second_upper_estimate": 0.36504977898166335
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29014.518308466722,
+      "required_pages_per_second_upper_estimate": 0.3721670874984508
     }
   ],
   "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": 0.16615242753903203,
-    "producer_p95_seconds": 0.27623364799728733,
-    "gpu_update_seconds": 0.9714808179853669,
-    "producer_batches_per_second": 6.018569904824791,
-    "gpu_batches_per_second": 1.0293564026037854,
-    "producer_headroom": 5.8469252142413,
+    "producer_batch_seconds": 0.20186159441419704,
+    "producer_p95_seconds": 0.2779740620026132,
+    "gpu_update_seconds": 0.952902257623627,
+    "producer_batches_per_second": 4.953889336413908,
+    "gpu_batches_per_second": 1.0494255753930384,
+    "producer_headroom": 4.720572332686424,
     "sustainable": true,
-    "buffer_seconds": 60.23181071509275,
+    "buffer_seconds": 60.0328422302885,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330557.9798914276,
-    "upload_bytes_per_second": 3766184.6559095164,
+    "generation_bytes_per_second": 371139.2872756181,
+    "upload_bytes_per_second": 3817756.4161513154,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.93004748875734,
+    "estimated_full_shards_per_hour": 79.63785136891754,
     "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": 0.8532274739991408,
-      "bytes": 1099982,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.0939723810006399,
       "bytes": 5792220,
@@ -562,6 +556,12 @@
       "bytes": 6102982,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.201682383001753,
+      "bytes": 8258960,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -221,3 +221,4 @@
 {"step": 667100, "loss": 0.5123287439346313, "lr": 0.0001, "grad_norm": 0.19962370620755204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97646820828, "tokens_per_second": 153365.6932093714, "data_wait_s": 1.3494001905201003e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 667200, "loss": 0.5280022025108337, "lr": 0.0001, "grad_norm": 0.20937009414418095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97661467110, "tokens_per_second": 152925.56086785797, "data_wait_s": 1.0196999937761575e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 667300, "loss": 0.5384029746055603, "lr": 0.0001, "grad_norm": 0.18967739220816573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97675919414, "tokens_per_second": 150362.78483535643, "data_wait_s": 2.0019000658066943e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 667400, "loss": 0.6393203735351562, "lr": 0.0001, "grad_norm": 0.21228044007224323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97690416559, "tokens_per_second": 149932.6551088959, "data_wait_s": 1.2204000086057931e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6e459b36abbb8963c36b64fb1d93adcc2a12f3affdf46796460aa18b1a65a8e
-size 475611
+oid sha256:09a9a74d4db84e8f62d8fc19ff9f29015069ddaa3f7fbe8b59f877f8e8b264ee
+size 477729