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": 75444.2244926276,
-      "required_pages_per_second_upper_estimate": 0.006539314167382126
+      "required_token_positions_per_second": 73173.16124402729,
+      "required_pages_per_second_upper_estimate": 0.00634246416095063
     },
     {
       "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": 9430.52806157845,
-      "required_pages_per_second_upper_estimate": 0.0037619786427231733
+      "required_token_positions_per_second": 9146.645155503411,
+      "required_pages_per_second_upper_estimate": 0.003648733507062155
     },
     {
       "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": 18861.0561231569,
-      "required_pages_per_second_upper_estimate": 0.004363173390841851
+      "required_token_positions_per_second": 18293.290311006822,
+      "required_pages_per_second_upper_estimate": 0.004231830762537943
     },
     {
       "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": 28291.58418473535,
-      "required_pages_per_second_upper_estimate": 0.001722809042379786
+      "required_token_positions_per_second": 27439.935466510233,
+      "required_pages_per_second_upper_estimate": 0.0016709481036953756
     },
     {
       "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": 28291.58418473535,
-      "required_pages_per_second_upper_estimate": 0.002177964466658405
+      "required_token_positions_per_second": 27439.935466510233,
+      "required_pages_per_second_upper_estimate": 0.0021124021908148958
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28291.58418473535,
-      "required_pages_per_second_upper_estimate": 0.3512867897331084
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27439.935466510233,
+      "required_pages_per_second_upper_estimate": 0.3407121629376589
     }
   ],
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.0796617153359307,
-    "producer_p95_seconds": 0.14203214499980277,
-    "gpu_update_seconds": 0.9772517445282334,
-    "producer_batches_per_second": 12.553081436710654,
-    "gpu_batches_per_second": 1.0232777844594674,
-    "producer_headroom": 12.26752073323047,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.58960816075047,
+    "buffer_seconds": 60.454952673799916,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 117,
+    "upload_pages_per_commit": 114,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 748388.808785961,
+    "generation_bytes_per_second": 665561.0215700947,
     "upload_bytes_per_second": 3893485.5279296017,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.2934083828169,
+    "estimated_full_shards_per_hour": 71.40695088065686,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -6,3 +6,4 @@
 {"step": 237700, "loss": 0.6223281361162663, "lr": 0.0003, "grad_norm": 0.16674797416904905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34791945903, "tokens_per_second": 148512.48516483826, "data_wait_s": 8.92599996404897e-06, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:2d47098ff164d387bdd5e2b2f9d6d026baf9c3b9969eed8a260b4e8357c698c1
-size 17109
+oid sha256:fce192059d6d176aeb49c007cb3b861fab6e7b656d464eb4db1123dba00b12d9
+size 19227