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": 73173.16124402729,
-      "required_pages_per_second_upper_estimate": 0.00634246416095063
+      "required_token_positions_per_second": 74550.87935475213,
+      "required_pages_per_second_upper_estimate": 0.006461881274993632
     },
     {
       "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": 9146.645155503411,
-      "required_pages_per_second_upper_estimate": 0.003648733507062155
+      "required_token_positions_per_second": 9318.859919344017,
+      "required_pages_per_second_upper_estimate": 0.003717432551198347
     },
     {
       "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": 18293.290311006822,
-      "required_pages_per_second_upper_estimate": 0.004231830762537943
+      "required_token_positions_per_second": 18637.719838688034,
+      "required_pages_per_second_upper_estimate": 0.004311508472014339
     },
     {
       "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": 27439.935466510233,
-      "required_pages_per_second_upper_estimate": 0.0016709481036953756
+      "required_token_positions_per_second": 27956.57975803205,
+      "required_pages_per_second_upper_estimate": 0.0017024090304259447
     },
     {
       "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": 27439.935466510233,
-      "required_pages_per_second_upper_estimate": 0.0021124021908148958
+      "required_token_positions_per_second": 27956.57975803205,
+      "required_pages_per_second_upper_estimate": 0.002152174898539179
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27439.935466510233,
-      "required_pages_per_second_upper_estimate": 0.3407121629376589
+      "required_token_positions_per_second": 27956.57975803205,
+      "required_pages_per_second_upper_estimate": 0.3471271559411457
     }
   ],
   "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": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.20986027589842493,
-    "producer_p95_seconds": 0.147974781999892,
-    "gpu_update_seconds": 1.007582544563332,
-    "producer_batches_per_second": 4.765075218351533,
-    "gpu_batches_per_second": 0.9924745177412555,
-    "producer_headroom": 4.801206613542312,
+    "producer_batch_seconds": 0.061685739398438244,
+    "producer_p95_seconds": 0.14232837400004428,
+    "gpu_update_seconds": 0.9889621777519693,
+    "producer_batches_per_second": 16.211202293302136,
+    "gpu_batches_per_second": 1.0111610155538213,
+    "producer_headroom": 16.0322659239618,
     "sustainable": true,
-    "buffer_seconds": 60.454952673799916,
+    "buffer_seconds": 60.32669284287012,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 114,
+    "upload_pages_per_commit": 116,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 665561.0215700947,
-    "upload_bytes_per_second": 3893485.5279296017,
+    "generation_bytes_per_second": 704396.7825034946,
+    "upload_bytes_per_second": 4582731.490574129,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.40695088065686,
+    "estimated_full_shards_per_hour": 75.573576003688,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 7655201,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1608069779999823,
+      "bytes": 11166576,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 237800, "loss": 0.5887066777795553, "lr": 0.0003, "grad_norm": 0.14764585941822073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34806607106, "tokens_per_second": 149644.0705802614, "data_wait_s": 1.3936999948782614e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 237900, "loss": 0.5877292566001415, "lr": 0.0003, "grad_norm": 0.1419477175668119, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34821316403, "tokens_per_second": 149208.26719476594, "data_wait_s": 1.9368000039321487e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 238000, "loss": 0.5937279835343361, "lr": 0.0003, "grad_norm": 0.1549946949955389, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34835903801, "tokens_per_second": 147951.72489235122, "data_wait_s": 1.2591000086104032e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 238100, "loss": 0.6234338097274303, "lr": 0.0003, "grad_norm": 0.15404151753059772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34850476923, "tokens_per_second": 147854.77742765663, "data_wait_s": 1.6987999970297096e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fce192059d6d176aeb49c007cb3b861fab6e7b656d464eb4db1123dba00b12d9
-size 19227
+oid sha256:c75395a6b4827f69fcf4b138bfb8c37d3ca6bc41034ae42a6ac3090a5a32356f
+size 21345