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": 76093.86079586418,
-      "required_pages_per_second_upper_estimate": 0.005264349851940438
+      "required_token_positions_per_second": 76996.27477643364,
+      "required_pages_per_second_upper_estimate": 0.00532678094500515
     },
     {
       "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": 9511.732599483023,
-      "required_pages_per_second_upper_estimate": 0.0039823638118401715
+      "required_token_positions_per_second": 9624.534347054205,
+      "required_pages_per_second_upper_estimate": 0.004029591547979875
     },
     {
       "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": 19023.465198966045,
-      "required_pages_per_second_upper_estimate": 0.003962086118430643
+      "required_token_positions_per_second": 19249.06869410841,
+      "required_pages_per_second_upper_estimate": 0.004009073376904539
     },
     {
       "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": 28535.197798449066,
-      "required_pages_per_second_upper_estimate": 0.00171064153729281
+      "required_token_positions_per_second": 28873.603041162616,
+      "required_pages_per_second_upper_estimate": 0.0017309284148786005
     },
     {
       "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": 28535.197798449066,
-      "required_pages_per_second_upper_estimate": 0.0022184861157589298
+      "required_token_positions_per_second": 28873.603041162616,
+      "required_pages_per_second_upper_estimate": 0.002244795634892554
     },
     {
       "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": 28535.197798449066,
-      "required_pages_per_second_upper_estimate": 0.36601887865021054
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28873.603041162616,
+      "required_pages_per_second_upper_estimate": 0.37035957775249956
     }
   ],
   "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": 0.10008997103889783,
-    "producer_p95_seconds": 0.2796466079998936,
-    "gpu_update_seconds": 0.9689086508278107,
-    "producer_batches_per_second": 9.991010983621639,
-    "gpu_batches_per_second": 1.0320890407425154,
-    "producer_headroom": 9.68037697254668,
+    "producer_batch_seconds": 0.13549749414065104,
+    "producer_p95_seconds": 0.28001820699864766,
+    "gpu_update_seconds": 0.9575528194588192,
+    "producer_batches_per_second": 7.380210286117658,
+    "gpu_batches_per_second": 1.044328813699458,
+    "producer_headroom": 7.066941167670943,
     "sustainable": true,
-    "buffer_seconds": 60.07233635132427,
+    "buffer_seconds": 60.32582762590561,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 271429.2600147634,
+    "generation_bytes_per_second": 334551.5039854984,
     "upload_bytes_per_second": 4680587.650313062,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.2424006493776,
+    "estimated_full_shards_per_hour": 71.78696479486193,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 915900, "loss": 0.5380746722221375, "lr": 3e-05, "grad_norm": 0.3630913997461515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134061298363, "tokens_per_second": 153389.72068559763, "data_wait_s": 1.2881999282399192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 916000, "loss": 0.5280999541282654, "lr": 3e-05, "grad_norm": 0.37965744720258804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134076062793, "tokens_per_second": 153416.2725599378, "data_wait_s": 1.3615997886518016e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 916100, "loss": 0.5013124942779541, "lr": 3e-05, "grad_norm": 0.34189349065592867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134090689795, "tokens_per_second": 152630.37685512129, "data_wait_s": 1.7843998648459092e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 916200, "loss": 0.5289828777313232, "lr": 3e-05, "grad_norm": 0.3310476315436398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134105152625, "tokens_per_second": 151393.85946864093, "data_wait_s": 7.460999768227339e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aefed66d673161530fd2f35b1458f0eaf30e20224cb416313756ad08d7b19ac6
-size 663561
+oid sha256:77ce1b792d0880eec5c556ee316a83844d32abc2d90fa848b2fb1962080497da
+size 665679