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": 74776.44346701661,
-      "required_pages_per_second_upper_estimate": 0.006481432600557701
+      "required_token_positions_per_second": 75437.1681151265,
+      "required_pages_per_second_upper_estimate": 0.006538702538464555
     },
     {
       "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": 9347.055433377076,
-      "required_pages_per_second_upper_estimate": 0.0037286801633066363
+      "required_token_positions_per_second": 9429.646014390812,
+      "required_pages_per_second_upper_estimate": 0.0037616267809122434
     },
     {
       "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": 18694.110866754152,
-      "required_pages_per_second_upper_estimate": 0.004324553543909238
+      "required_token_positions_per_second": 18859.292028781623,
+      "required_pages_per_second_upper_estimate": 0.004362765298655131
     },
     {
       "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": 28041.16630013123,
-      "required_pages_per_second_upper_estimate": 0.0017075599070484964
+      "required_token_positions_per_second": 28288.938043172435,
+      "required_pages_per_second_upper_estimate": 0.0017226479062418362
     },
     {
       "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": 28041.16630013123,
-      "required_pages_per_second_upper_estimate": 0.0021586866047005084
+      "required_token_positions_per_second": 28288.938043172435,
+      "required_pages_per_second_upper_estimate": 0.002177760759355903
     },
     {
       "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": 28041.16630013123,
-      "required_pages_per_second_upper_estimate": 0.3481774377010719
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28288.938043172435,
+      "required_pages_per_second_upper_estimate": 0.3512539335109631
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.057545325234315214,
-    "producer_p95_seconds": 0.19222766600069008,
-    "gpu_update_seconds": 0.9859789605067394,
-    "producer_batches_per_second": 17.377606190045196,
-    "gpu_batches_per_second": 1.014220424628589,
-    "producer_headroom": 17.13395408735624,
+    "producer_batch_seconds": 0.4773819408281099,
+    "producer_p95_seconds": 0.24347419500008982,
+    "gpu_update_seconds": 0.9773431564594512,
+    "producer_batches_per_second": 2.0947587549401416,
+    "gpu_batches_per_second": 1.0231820762142807,
+    "producer_headroom": 2.047298133574268,
     "sustainable": true,
-    "buffer_seconds": 60.1447165909111,
+    "buffer_seconds": 60.59527570048598,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 233500.85525451138,
-    "upload_bytes_per_second": 6474041.960848427,
+    "generation_bytes_per_second": 220519.2721500514,
+    "upload_bytes_per_second": 6479780.094583534,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.10384791590219,
+    "estimated_full_shards_per_hour": 47.318302377473415,
     "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": 1.2178119280006285,
-      "bytes": 3672261,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.0616448490000039,
       "bytes": 5911843,
@@ -562,6 +556,12 @@
       "bytes": 8839899,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 0.756701359001454,
+      "bytes": 1071685,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 11000, "loss": 0.722653703764081, "lr": 0.0003, "grad_norm": 0.14722297485559593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1609008434, "tokens_per_second": 139375.47298112747, "data_wait_s": 8.352999429916963e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 11100, "loss": 0.7419378254562616, "lr": 0.0003, "grad_norm": 0.15071296947925078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1623604035, "tokens_per_second": 148108.5150613146, "data_wait_s": 1.1514999641804025e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 11200, "loss": 0.7446037046611309, "lr": 0.0003, "grad_norm": 0.14735876864574618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1638338285, "tokens_per_second": 148852.62966130654, "data_wait_s": 1.3658000170835294e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 11300, "loss": 0.6859790533781052, "lr": 0.0003, "grad_norm": 0.16650099415743275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1653152694, "tokens_per_second": 150670.7353186621, "data_wait_s": 1.068199890141841e-05, "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:a4c8ad4db1b998516e3d98ec68748273326f1207ba6e6e75823f62851130fcc0
-size 234968
+oid sha256:eaeaca272cb438affa7ec069137d6288f5c728d257cd50591a9d4c74b6963fad
+size 237051