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": 77439.95617235676,
-      "required_pages_per_second_upper_estimate": 0.005357475853457785
+      "required_token_positions_per_second": 77077.52393358367,
+      "required_pages_per_second_upper_estimate": 0.005332401950220807
     },
     {
       "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": 9679.994521544595,
-      "required_pages_per_second_upper_estimate": 0.004052811564898862
+      "required_token_positions_per_second": 9634.690491697958,
+      "required_pages_per_second_upper_estimate": 0.004033843713657797
     },
     {
       "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": 19359.98904308919,
-      "required_pages_per_second_upper_estimate": 0.0040321751604309244
+      "required_token_positions_per_second": 19269.380983395917,
+      "required_pages_per_second_upper_estimate": 0.004013303891092036
     },
     {
       "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": 29039.983564633785,
-      "required_pages_per_second_upper_estimate": 0.0017409026732123475
+      "required_token_positions_per_second": 28904.07147509387,
+      "required_pages_per_second_upper_estimate": 0.0017327549509701743
     },
     {
       "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": 29039.983564633785,
-      "required_pages_per_second_upper_estimate": 0.0022577309887618566
+      "required_token_positions_per_second": 28904.07147509387,
+      "required_pages_per_second_upper_estimate": 0.0022471644216142312
     },
     {
       "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": 29039.983564633785,
-      "required_pages_per_second_upper_estimate": 0.3724937284621001
+      "required_token_positions_per_second": 28904.07147509387,
+      "required_pages_per_second_upper_estimate": 0.3707503941085141
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09047138513292907,
-    "producer_p95_seconds": 0.25487454799804254,
-    "gpu_update_seconds": 0.9520666545304451,
-    "producer_batches_per_second": 11.053218634054359,
-    "gpu_batches_per_second": 1.050346627771766,
-    "producer_headroom": 10.52340088671771,
+    "producer_batch_seconds": 0.11722676057024728,
+    "producer_p95_seconds": 0.2510532750020502,
+    "gpu_update_seconds": 0.9565434414256757,
+    "producer_batches_per_second": 8.530475423320746,
+    "gpu_batches_per_second": 1.0454308259220875,
+    "producer_headroom": 8.159770318420374,
     "sustainable": true,
-    "buffer_seconds": 60.93226588994849,
+    "buffer_seconds": 60.26223680981757,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343134.03537322563,
+    "generation_bytes_per_second": 328884.25548470847,
     "upload_bytes_per_second": 5625859.543825907,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.62857623956276,
+    "estimated_full_shards_per_hour": 70.57090519338551,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -233,3 +233,4 @@
 {"step": 708300, "loss": 0.5177313685417175, "lr": 8e-05, "grad_norm": 0.2094146281548363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103675045922, "tokens_per_second": 151721.92447165836, "data_wait_s": 1.831500048865564e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 708400, "loss": 0.5601287484169006, "lr": 8e-05, "grad_norm": 0.21821455588675287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103689664989, "tokens_per_second": 152421.051121285, "data_wait_s": 1.931499718921259e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 708500, "loss": 0.5990826487541199, "lr": 8e-05, "grad_norm": 0.21880586457356885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103704404070, "tokens_per_second": 153765.63286826626, "data_wait_s": 1.557999712531455e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 708600, "loss": 0.5554506182670593, "lr": 8e-05, "grad_norm": 0.22612946778818444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103719089001, "tokens_per_second": 137928.13761065842, "data_wait_s": 1.6326001059496775e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d5d951170c43f8008d2f1df946479a8cece995792a06c962c986153e5a6f915
-size 501027
+oid sha256:37cbfe846827037eeb51ba692a40edfadbf44b583af5819e7e97a07ca112e0cc
+size 503145