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": 74770.46689945106,
-      "required_pages_per_second_upper_estimate": 0.006480914566828591
+      "required_token_positions_per_second": 75319.69095557561,
+      "required_pages_per_second_upper_estimate": 0.006528519916017823
     },
     {
       "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": 9346.308362431382,
-      "required_pages_per_second_upper_estimate": 0.003728382145536693
+      "required_token_positions_per_second": 9414.961369446952,
+      "required_pages_per_second_upper_estimate": 0.00375576885648913
     },
     {
       "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": 18692.616724862764,
-      "required_pages_per_second_upper_estimate": 0.004324207900478665
+      "required_token_positions_per_second": 18829.922738893903,
+      "required_pages_per_second_upper_estimate": 0.004355971230321456
     },
     {
       "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": 28038.925087294145,
-      "required_pages_per_second_upper_estimate": 0.001707423428945453
+      "required_token_positions_per_second": 28244.884108340855,
+      "required_pages_per_second_upper_estimate": 0.001719965252743726
     },
     {
       "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": 28038.925087294145,
-      "required_pages_per_second_upper_estimate": 0.0021585140699322356
+      "required_token_positions_per_second": 28244.884108340855,
+      "required_pages_per_second_upper_estimate": 0.0021743693655034715
     },
     {
       "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": 28038.925087294145,
-      "required_pages_per_second_upper_estimate": 0.3481496093384922
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28244.884108340855,
+      "required_pages_per_second_upper_estimate": 0.35070693107939027
     }
   ],
   "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": 0.3627634417499763,
-    "producer_p95_seconds": 0.3269775070002652,
-    "gpu_update_seconds": 0.9860577719696076,
-    "producer_batches_per_second": 2.756617356964045,
-    "gpu_batches_per_second": 1.0141393622429884,
-    "producer_headroom": 2.7181839691807146,
+    "producer_batch_seconds": 0.06431004463289014,
+    "producer_p95_seconds": 0.21595620100015367,
+    "gpu_update_seconds": 0.9788675320439856,
+    "producer_batches_per_second": 15.549670439640298,
+    "gpu_batches_per_second": 1.0215886902611708,
+    "producer_headroom": 15.221067527348014,
     "sustainable": true,
-    "buffer_seconds": 60.149524090146066,
+    "buffer_seconds": 60.68978698672711,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339490.2539706306,
+    "generation_bytes_per_second": 339517.6709511822,
     "upload_bytes_per_second": 5191784.548656985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.8467055734557,
+    "estimated_full_shards_per_hour": 72.85258861924743,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 209900, "loss": 0.6171294096857309, "lr": 0.0003, "grad_norm": 0.14364944481939032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30721359198, "tokens_per_second": 146652.94399301035, "data_wait_s": 1.898600021377206e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 210000, "loss": 0.5918615683913231, "lr": 0.0003, "grad_norm": 0.17345168380647322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30736062273, "tokens_per_second": 148821.90664461232, "data_wait_s": 1.5919998986646533e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 210100, "loss": 0.5807416588068008, "lr": 0.0003, "grad_norm": 0.1436917269689581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30750746073, "tokens_per_second": 125072.04918072904, "data_wait_s": 1.2976999641978182e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 210200, "loss": 0.5917802564799786, "lr": 0.0003, "grad_norm": 0.16181771319320384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30765397393, "tokens_per_second": 144146.832485573, "data_wait_s": 1.1764999726437964e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:220a0b1f625c5de9173777af3efd8a7192c65fe3b511ebbd8b29e6a407eaee76
-size 322332
+oid sha256:906525246047175ec8516cbb10288be25bd269250a7cda83e513ce9a26850f9b
+size 324450