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": 74683.09873289081,
-      "required_pages_per_second_upper_estimate": 0.006473341715583742
+      "required_token_positions_per_second": 75376.17064754231,
+      "required_pages_per_second_upper_estimate": 0.006533415432571027
     },
     {
       "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": 9335.387341611351,
-      "required_pages_per_second_upper_estimate": 0.0037240255870477704
+      "required_token_positions_per_second": 9422.021330942789,
+      "required_pages_per_second_upper_estimate": 0.0037585851806856505
     },
     {
       "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": 18670.774683222702,
-      "required_pages_per_second_upper_estimate": 0.00431915512855204
+      "required_token_positions_per_second": 18844.042661885578,
+      "required_pages_per_second_upper_estimate": 0.004359237626003422
     },
     {
       "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": 28006.16202483405,
-      "required_pages_per_second_upper_estimate": 0.0017054283303361373
+      "required_token_positions_per_second": 28266.063992828367,
+      "required_pages_per_second_upper_estimate": 0.0017212549965867497
     },
     {
       "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": 28006.16202483405,
-      "required_pages_per_second_upper_estimate": 0.002155991878690096
+      "required_token_positions_per_second": 28266.063992828367,
+      "required_pages_per_second_upper_estimate": 0.002175999851640993
     },
     {
       "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": 28006.16202483405,
-      "required_pages_per_second_upper_estimate": 0.34774280175365424
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28266.063992828367,
+      "required_pages_per_second_upper_estimate": 0.3509699143602117
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.08516968315609574,
-    "producer_p95_seconds": 0.25121986500016646,
-    "gpu_update_seconds": 0.9872113135489091,
-    "producer_batches_per_second": 11.741267114581584,
-    "gpu_batches_per_second": 1.0129543556435927,
-    "producer_headroom": 11.591111730914696,
+    "producer_batch_seconds": 0.097038341445284,
+    "producer_p95_seconds": 0.23142350999842165,
+    "gpu_update_seconds": 0.9781340623517593,
+    "producer_batches_per_second": 10.30520498501986,
+    "gpu_batches_per_second": 1.0223547451109798,
+    "producer_headroom": 10.079872015365076,
     "sustainable": true,
-    "buffer_seconds": 60.219890126483456,
+    "buffer_seconds": 60.64431186580907,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355392.05764273176,
+    "generation_bytes_per_second": 355421.5343195955,
     "upload_bytes_per_second": 4847801.931672915,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.25886246644463,
+    "estimated_full_shards_per_hour": 76.26518747511767,
     "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

@@ -167,3 +167,4 @@
 {"step": 253800, "loss": 0.5735029503703117, "lr": 0.0003, "grad_norm": 0.14453678746875576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37149773881, "tokens_per_second": 151450.40581484855, "data_wait_s": 1.007000173558481e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 253900, "loss": 0.5994154773652554, "lr": 0.0003, "grad_norm": 0.15160254849164415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37164437668, "tokens_per_second": 141735.944533427, "data_wait_s": 1.719699866953306e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 254000, "loss": 0.5805207639932632, "lr": 0.0003, "grad_norm": 0.14634467205067897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37179037470, "tokens_per_second": 147658.0662841845, "data_wait_s": 1.003299985313788e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 254100, "loss": 0.5910433381795883, "lr": 0.0003, "grad_norm": 0.15330688472435477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37193591900, "tokens_per_second": 148962.12180058032, "data_wait_s": 1.4354001905303448e-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:f6fcdb5cda4cffa40ae306969e0c3590759bea55cf53c601693a6a87c8df2ca4
-size 360456
+oid sha256:b84f52809c0065103ed0ed89efee8b4fdfa354312bda1adccb784238cfe91e8d
+size 362574