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": 77375.53717432328,
-      "required_pages_per_second_upper_estimate": 0.005353019197701161
+      "required_token_positions_per_second": 77425.78040687449,
+      "required_pages_per_second_upper_estimate": 0.005356495141109413
     },
     {
       "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": 9671.94214679041,
-      "required_pages_per_second_upper_estimate": 0.004049440203741991
+      "required_token_positions_per_second": 9678.222550859311,
+      "required_pages_per_second_upper_estimate": 0.004052069677775889
     },
     {
       "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": 19343.88429358082,
-      "required_pages_per_second_upper_estimate": 0.004028820965819136
+      "required_token_positions_per_second": 19356.445101718622,
+      "required_pages_per_second_upper_estimate": 0.004031437050903454
     },
     {
       "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": 29015.82644037123,
-      "required_pages_per_second_upper_estimate": 0.0017394544904986006
+      "required_token_positions_per_second": 29034.66765257793,
+      "required_pages_per_second_upper_estimate": 0.0017405839924015268
     },
     {
       "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": 29015.82644037123,
-      "required_pages_per_second_upper_estimate": 0.0022558528786064034
+      "required_token_positions_per_second": 29034.66765257793,
+      "required_pages_per_second_upper_estimate": 0.0022573176999817397
     },
     {
       "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": 29015.82644037123,
-      "required_pages_per_second_upper_estimate": 0.37218386680995924
+      "required_token_positions_per_second": 29034.66765257793,
+      "required_pages_per_second_upper_estimate": 0.3724255416500292
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.0987689845156865,
-    "producer_p95_seconds": 0.293807119000121,
-    "gpu_update_seconds": 0.9528592975567257,
-    "producer_batches_per_second": 10.124635834857449,
-    "gpu_batches_per_second": 1.0494728891916678,
-    "producer_headroom": 9.647353389619921,
+    "producer_batch_seconds": 0.07751702610917732,
+    "producer_p95_seconds": 0.27455812900007004,
+    "gpu_update_seconds": 0.9522409669306199,
+    "producer_batches_per_second": 12.900391697064975,
+    "gpu_batches_per_second": 1.0501543566470606,
+    "producer_headroom": 12.284281463396892,
     "sustainable": true,
-    "buffer_seconds": 60.03013574607372,
+    "buffer_seconds": 60.94342188355967,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384507.2961762014,
-    "upload_bytes_per_second": 4819197.454597539,
+    "generation_bytes_per_second": 405879.39226201404,
+    "upload_bytes_per_second": 4935043.569591223,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.50631488766223,
+    "estimated_full_shards_per_hour": 87.0922691907436,
     "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": [
-    {
-      "seconds": 3.434384232998127,
-      "bytes": 5971898,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1304158860002644,
       "bytes": 8230136,
@@ -562,6 +556,12 @@
       "bytes": 2347731,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3670132119987102,
+      "bytes": 6121191,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -250,3 +250,4 @@
 {"step": 910000, "loss": 0.4874023199081421, "lr": 3e-05, "grad_norm": 0.31720257835153753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133197845917, "tokens_per_second": 152953.92772428304, "data_wait_s": 1.6943999071372673e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 910100, "loss": 0.5183731913566589, "lr": 3e-05, "grad_norm": 0.3192280041209559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133212291168, "tokens_per_second": 139551.42774045252, "data_wait_s": 1.3638000382343307e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 910200, "loss": 0.49870553612709045, "lr": 3e-05, "grad_norm": 0.3184701814368264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133226966404, "tokens_per_second": 152719.89098807538, "data_wait_s": 1.440599953639321e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 910300, "loss": 0.5121055245399475, "lr": 3e-05, "grad_norm": 0.34960591957401166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133241637469, "tokens_per_second": 152278.4671584382, "data_wait_s": 2.0385999960126355e-05, "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:fb64019b36f9c3951ca4ac61fa8811527261e8f3a13b745c9c3f223e31dfcd6d
-size 537816
+oid sha256:4dac5cbdb551c9e410227135632b5580726ac982e2b7e7f17b5f4edfc0f3cd8f
+size 539934