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": 76109.27578882316,
-      "required_pages_per_second_upper_estimate": 0.005265416296921051
+      "required_token_positions_per_second": 76514.9846693944,
+      "required_pages_per_second_upper_estimate": 0.0052934841786532695
     },
     {
       "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": 9513.659473602895,
-      "required_pages_per_second_upper_estimate": 0.003983170553796454
+      "required_token_positions_per_second": 9564.3730836743,
+      "required_pages_per_second_upper_estimate": 0.004004403283312748
     },
     {
       "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": 19027.31894720579,
-      "required_pages_per_second_upper_estimate": 0.0039628887525588056
+      "required_token_positions_per_second": 19128.7461673486,
+      "required_pages_per_second_upper_estimate": 0.003984013367698884
     },
     {
       "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": 28540.978420808686,
-      "required_pages_per_second_upper_estimate": 0.00171098807677677
+      "required_token_positions_per_second": 28693.1192510229,
+      "required_pages_per_second_upper_estimate": 0.0017201086872425154
     },
     {
       "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": 28540.978420808686,
-      "required_pages_per_second_upper_estimate": 0.0022189355337211194
+      "required_token_positions_per_second": 28693.1192510229,
+      "required_pages_per_second_upper_estimate": 0.0022307638140734855
     },
     {
       "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": 28540.978420808686,
-      "required_pages_per_second_upper_estimate": 0.36609302626709106
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28693.1192510229,
+      "required_pages_per_second_upper_estimate": 0.3680445254809828
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1872666168046777,
+    "producer_batch_seconds": 0.23772263537493643,
     "producer_p95_seconds": 0.2463551250002638,
-    "gpu_update_seconds": 0.9687124103580965,
-    "producer_batches_per_second": 5.3399800619189755,
-    "gpu_batches_per_second": 1.0322981199655918,
-    "producer_headroom": 5.172904957045708,
+    "gpu_update_seconds": 0.9635759625198071,
+    "producer_batches_per_second": 4.206583013951527,
+    "gpu_batches_per_second": 1.037800898836187,
+    "producer_headroom": 4.053362276587814,
     "sustainable": true,
-    "buffer_seconds": 60.06016944220198,
+    "buffer_seconds": 60.70528563874785,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 464063.7288726379,
-    "upload_bytes_per_second": 4696463.237457393,
+    "generation_bytes_per_second": 463237.2937729814,
+    "upload_bytes_per_second": 4741709.878562262,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.57727336534836,
+    "estimated_full_shards_per_hour": 99.39993963138657,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 6139992,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.280977876000179,
+      "bytes": 9975922,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -116,3 +116,4 @@
 {"step": 731600, "loss": 0.5479867458343506, "lr": 5e-05, "grad_norm": 0.2170199591214406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107085434730, "tokens_per_second": 151005.64977266974, "data_wait_s": 1.911899926199112e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 731700, "loss": 0.5008614659309387, "lr": 5e-05, "grad_norm": 0.21254040492607887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107100156922, "tokens_per_second": 151196.52390834645, "data_wait_s": 2.17869983316632e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 731800, "loss": 0.5290740132331848, "lr": 5e-05, "grad_norm": 0.22209223910479947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107114684526, "tokens_per_second": 149706.74671501736, "data_wait_s": 2.137299998139497e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 731900, "loss": 0.5578677654266357, "lr": 5e-05, "grad_norm": 0.2192783896648435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107129246802, "tokens_per_second": 142523.7032786384, "data_wait_s": 1.9320001229061745e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1aae7d16de9563f9de802dd5b5e455066aa7031c3e801ef08179bf6a2613002b
-size 251655
+oid sha256:c9f4d4f4c39d9376c395dde0209b7eaad5cf4f5713c6c4dea9a0097ddd4a7974
+size 253773