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": 74895.41714252319,
-      "required_pages_per_second_upper_estimate": 0.006491744937214588
+      "required_token_positions_per_second": 75530.40719772117,
+      "required_pages_per_second_upper_estimate": 0.006546784265831568
     },
     {
       "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": 9361.927142815399,
-      "required_pages_per_second_upper_estimate": 0.003734612710553454
+      "required_token_positions_per_second": 9441.300899715146,
+      "required_pages_per_second_upper_estimate": 0.0037662760889241847
     },
     {
       "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": 18723.854285630798,
-      "required_pages_per_second_upper_estimate": 0.004331434160400056
+      "required_token_positions_per_second": 18882.601799430293,
+      "required_pages_per_second_upper_estimate": 0.004368157603856749
     },
     {
       "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": 28085.781428446197,
-      "required_pages_per_second_upper_estimate": 0.00171027673428539
+      "required_token_positions_per_second": 28323.90269914544,
+      "required_pages_per_second_upper_estimate": 0.0017247770703452197
     },
     {
       "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": 28085.781428446197,
-      "required_pages_per_second_upper_estimate": 0.0021621212007807743
+      "required_token_positions_per_second": 28323.90269914544,
+      "required_pages_per_second_upper_estimate": 0.0021804524353610678
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28085.781428446197,
-      "required_pages_per_second_upper_estimate": 0.34873140827751464
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28323.90269914544,
+      "required_pages_per_second_upper_estimate": 0.3516880775189719
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11861088077353088,
-    "producer_p95_seconds": 0.3669769870029995,
-    "gpu_update_seconds": 0.9844127025782947,
-    "producer_batches_per_second": 8.430929721442212,
-    "gpu_batches_per_second": 1.015834108378407,
-    "producer_headroom": 8.299514312332596,
+    "producer_batch_seconds": 0.10826317090607063,
+    "producer_p95_seconds": 0.3732283249992179,
+    "gpu_update_seconds": 0.9761366678058164,
+    "producer_batches_per_second": 9.236751442164966,
+    "gpu_batches_per_second": 1.024446712208675,
+    "producer_headroom": 9.01633177410548,
     "sustainable": true,
-    "buffer_seconds": 60.04917485727598,
+    "buffer_seconds": 60.520473403960615,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430633.3026476258,
+    "generation_bytes_per_second": 447561.9602140484,
     "upload_bytes_per_second": 6264638.110340561,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.40388211795407,
+    "estimated_full_shards_per_hour": 96.03637795272911,
     "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

@@ -319,3 +319,4 @@
 {"step": 31900, "loss": 0.6768332421779633, "lr": 0.0003, "grad_norm": 0.12667614660089868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4668034813, "tokens_per_second": 149393.52081098716, "data_wait_s": 1.752900061546825e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 32000, "loss": 0.6774434354156256, "lr": 0.0003, "grad_norm": 0.15375056451475647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4682822912, "tokens_per_second": 150189.13341213146, "data_wait_s": 8.912997145671397e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 32100, "loss": 0.6099816374480724, "lr": 0.0003, "grad_norm": 0.14531868021638142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4697595341, "tokens_per_second": 150225.49777577218, "data_wait_s": 1.0828000085894018e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 32200, "loss": 0.6732552777975798, "lr": 0.0003, "grad_norm": 0.14508445797455272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4712325524, "tokens_per_second": 150656.4921985203, "data_wait_s": 9.315997886005789e-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:feba32c46702be4e682970b7dd9bc3d854ccef06dc567b1d5edf718d01f8e099
-size 678967
+oid sha256:d24f489761e24c6da531a759892a7961f10897daebb5aff6dac57b1d607706bd
+size 681085