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": 75682.08206380114,
-      "required_pages_per_second_upper_estimate": 0.006559931058806945
+      "required_token_positions_per_second": 74506.63456218303,
+      "required_pages_per_second_upper_estimate": 0.006458046248511143
     },
     {
       "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": 9460.260257975142,
-      "required_pages_per_second_upper_estimate": 0.003773839260401764
+      "required_token_positions_per_second": 9313.329320272878,
+      "required_pages_per_second_upper_estimate": 0.0037152263125390453
     },
     {
       "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": 18920.520515950284,
-      "required_pages_per_second_upper_estimate": 0.004376929431577031
+      "required_token_positions_per_second": 18626.658640545756,
+      "required_pages_per_second_upper_estimate": 0.004308949658494562
     },
     {
       "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": 28380.780773925428,
-      "required_pages_per_second_upper_estimate": 0.0017282406466831253
+      "required_token_positions_per_second": 27939.987960818635,
+      "required_pages_per_second_upper_estimate": 0.0017013986770260802
     },
     {
       "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": 28380.780773925428,
-      "required_pages_per_second_upper_estimate": 0.002184831067006219
+      "required_token_positions_per_second": 27939.987960818635,
+      "required_pages_per_second_upper_estimate": 0.0021508976160606558
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28380.780773925428,
-      "required_pages_per_second_upper_estimate": 0.35239431284906847
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27939.987960818635,
+      "required_pages_per_second_upper_estimate": 0.34692114134892826
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10271027421100598,
-    "producer_p95_seconds": 0.3439941659999022,
-    "gpu_update_seconds": 0.9741803870808705,
-    "producer_batches_per_second": 9.73612433304987,
-    "gpu_batches_per_second": 1.0265039342420945,
-    "producer_headroom": 9.484741371438005,
+    "producer_batch_seconds": 0.26439618542971743,
+    "producer_p95_seconds": 0.34165482800017344,
+    "gpu_update_seconds": 0.9895494600345532,
+    "producer_batches_per_second": 3.782202827074534,
+    "gpu_batches_per_second": 1.0105609071476649,
+    "producer_headroom": 3.742676765272766,
     "sustainable": true,
-    "buffer_seconds": 60.39918399901397,
+    "buffer_seconds": 60.36251706210775,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 464126.9001165235,
+    "generation_bytes_per_second": 438080.54928837816,
     "upload_bytes_per_second": 5696513.289361367,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.59082844373492,
+    "estimated_full_shards_per_hour": 94.00188788403042,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 24200, "loss": 0.6809567473828793, "lr": 0.0003, "grad_norm": 0.1368268725427712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3541118262, "tokens_per_second": 149641.55170525573, "data_wait_s": 1.2199001503176987e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 24300, "loss": 0.6602495796978474, "lr": 0.0003, "grad_norm": 0.1262718785820121, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3555844316, "tokens_per_second": 149957.1482101085, "data_wait_s": 1.751700256136246e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 24400, "loss": 0.6389599852263927, "lr": 0.0003, "grad_norm": 0.13699569618460566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3570585881, "tokens_per_second": 150419.3254969971, "data_wait_s": 8.304999937536195e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 24500, "loss": 0.6439226306974888, "lr": 0.0003, "grad_norm": 0.13887686574509203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3585136027, "tokens_per_second": 147514.78263566113, "data_wait_s": 8.819999493425712e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e44217e9a560caef76ff91fa31dacd2bb8018a6a118e8dff76386b6f9a8de4d4
-size 514315
+oid sha256:8785d828ee03be79ac30e36591f234fe260ea30a8886f0e2cbd849e13ae75b14
+size 516433