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": 76906.71322483788,
-      "required_pages_per_second_upper_estimate": 0.005320584869054325
+      "required_token_positions_per_second": 77544.64390745682,
+      "required_pages_per_second_upper_estimate": 0.005364718393881019
     },
     {
       "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": 9613.339153104735,
-      "required_pages_per_second_upper_estimate": 0.004024904354055466
+      "required_token_positions_per_second": 9693.080488432102,
+      "required_pages_per_second_upper_estimate": 0.004058290385968598
     },
     {
       "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": 19226.67830620947,
-      "required_pages_per_second_upper_estimate": 0.004004410049579426
+      "required_token_positions_per_second": 19386.160976864205,
+      "required_pages_per_second_upper_estimate": 0.0040376260840359526
     },
     {
       "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": 28840.017459314207,
-      "required_pages_per_second_upper_estimate": 0.0017289150105290025
+      "required_token_positions_per_second": 29079.24146529631,
+      "required_pages_per_second_upper_estimate": 0.0017432561244137225
     },
     {
       "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": 28840.017459314207,
-      "required_pages_per_second_upper_estimate": 0.002242184503631202
+      "required_token_positions_per_second": 29079.24146529631,
+      "required_pages_per_second_upper_estimate": 0.002260783117861112
     },
     {
       "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": 28840.017459314207,
-      "required_pages_per_second_upper_estimate": 0.36992877796993634
+      "required_token_positions_per_second": 29079.24146529631,
+      "required_pages_per_second_upper_estimate": 0.3729972866599493
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05902043485926356,
-    "producer_p95_seconds": 0.19171496400122123,
-    "gpu_update_seconds": 0.958667935586522,
-    "producer_batches_per_second": 16.943284175803473,
-    "gpu_batches_per_second": 1.043114057411538,
-    "producer_headroom": 16.242983262873302,
+    "producer_batch_seconds": 0.08051996124983418,
+    "producer_p95_seconds": 0.2062423770003079,
+    "gpu_update_seconds": 0.950781334272272,
+    "producer_batches_per_second": 12.41928069112253,
+    "gpu_batches_per_second": 1.0517665460538306,
+    "producer_headroom": 11.808020266207343,
     "sustainable": true,
-    "buffer_seconds": 60.39607994195089,
+    "buffer_seconds": 60.850005393425405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297470.26558166224,
-    "upload_bytes_per_second": 3275717.099881571,
+    "generation_bytes_per_second": 317803.5448677939,
+    "upload_bytes_per_second": 3292210.839553986,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.83019424044991,
+    "estimated_full_shards_per_hour": 68.19324264073718,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 11662593,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2687110730003042,
+      "bytes": 6191110,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -115,3 +115,4 @@
 {"step": 696500, "loss": 0.5282445549964905, "lr": 8e-05, "grad_norm": 0.2143935001575543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101949046168, "tokens_per_second": 153406.3899245804, "data_wait_s": 1.0687999747460708e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 696600, "loss": 0.5328052043914795, "lr": 8e-05, "grad_norm": 0.20013354791740276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101963504039, "tokens_per_second": 150144.31962397718, "data_wait_s": 1.7787000615498982e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 696700, "loss": 0.5734730362892151, "lr": 8e-05, "grad_norm": 0.22531892352341243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101978008930, "tokens_per_second": 151518.06675428574, "data_wait_s": 2.0571000277413987e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 696800, "loss": 0.6192939877510071, "lr": 8e-05, "grad_norm": 0.21230961375212332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101992517449, "tokens_per_second": 150895.99518598872, "data_wait_s": 1.5876999896136113e-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 sha256:6250bbb6e578bbdd89cec03e644c21638d1e43cc291f738bb0b8d5268a3cd5fe
-size 249537
+oid sha256:24311d11428c54252710333fbf16e0e68d51e205a506a2fe19bb0831d39cd799
+size 251655