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": 74188.10494249329,
-      "required_pages_per_second_upper_estimate": 0.0064304369083823774
+      "required_token_positions_per_second": 75444.50302284452,
+      "required_pages_per_second_upper_estimate": 0.0065393383096754
     },
     {
       "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": 9273.51311781166,
-      "required_pages_per_second_upper_estimate": 0.0036993430340719886
+      "required_token_positions_per_second": 9430.562877855566,
+      "required_pages_per_second_upper_estimate": 0.0037619925314566643
     },
     {
       "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": 18547.02623562332,
-      "required_pages_per_second_upper_estimate": 0.004290528103098217
+      "required_token_positions_per_second": 18861.12575571113,
+      "required_pages_per_second_upper_estimate": 0.004363189499107781
     },
     {
       "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": 27820.539353434982,
-      "required_pages_per_second_upper_estimate": 0.001694124883534825
+      "required_token_positions_per_second": 28291.688633566697,
+      "required_pages_per_second_upper_estimate": 0.0017228154027656142
     },
     {
       "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": 27820.539353434982,
-      "required_pages_per_second_upper_estimate": 0.002141702131609354
+      "required_token_positions_per_second": 28291.688633566697,
+      "required_pages_per_second_upper_estimate": 0.0021779725074185753
     },
     {
       "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": 27820.539353434982,
-      "required_pages_per_second_upper_estimate": 0.34543798941399584
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28291.688633566697,
+      "required_pages_per_second_upper_estimate": 0.35128808663802596
     }
   ],
   "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": 0.0972503262421327,
-    "producer_p95_seconds": 0.2202963540003111,
-    "gpu_update_seconds": 0.9937981305379086,
-    "producer_batches_per_second": 10.282741854358534,
-    "gpu_batches_per_second": 1.0062405726792167,
-    "producer_headroom": 10.218969631665418,
+    "producer_batch_seconds": 0.27071476767979163,
+    "producer_p95_seconds": 0.2606421410000621,
+    "gpu_update_seconds": 0.9772481366558307,
+    "producer_batches_per_second": 3.693924821947008,
+    "gpu_batches_per_second": 1.0232815622673141,
+    "producer_headroom": 3.6098811491944347,
     "sustainable": true,
-    "buffer_seconds": 60.62168596281242,
+    "buffer_seconds": 60.5893844726615,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 413502.0198293568,
-    "upload_bytes_per_second": 5970183.370832096,
+    "generation_bytes_per_second": 394144.88152834936,
+    "upload_bytes_per_second": 5904366.182918268,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.72790762100723,
+    "estimated_full_shards_per_hour": 84.57431635272847,
     "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": [
@@ -544,6 +544,12 @@
       "bytes": 6032233,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1764326539996546,
+      "bytes": 2559639,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -121,3 +121,4 @@
 {"step": 127100, "loss": 0.5815989589318633, "lr": 0.0003, "grad_norm": 0.13598667592178712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18600543329, "tokens_per_second": 149133.4866198227, "data_wait_s": 1.9907000023522414e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 127200, "loss": 0.6274984590709209, "lr": 0.0003, "grad_norm": 0.13884664794655535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18615248565, "tokens_per_second": 149852.5159162955, "data_wait_s": 2.2731999706593342e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 127300, "loss": 0.6228926777839661, "lr": 0.0003, "grad_norm": 0.14142022313523672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18629941283, "tokens_per_second": 150273.28461330765, "data_wait_s": 1.1657000868581235e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 127400, "loss": 0.6144789196550846, "lr": 0.0003, "grad_norm": 0.14685550382924611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18644665520, "tokens_per_second": 150107.67555962576, "data_wait_s": 1.3332999515114352e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee9fc5fa6ef89227dc277ca8e243cd8b1aa4724a68950d433ae9420a6004e6a3
-size 262245
+oid sha256:8a36a40a32b4edf1ab665cd725ceea369573fb8a4deb8cf76ac0c20f6e0b82ed
+size 264363