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": 75978.37675145088,
-      "required_pages_per_second_upper_estimate": 0.0065856131313804045
+      "required_token_positions_per_second": 74395.56145044182,
+      "required_pages_per_second_upper_estimate": 0.006448418712697644
     },
     {
       "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": 9497.29709393136,
-      "required_pages_per_second_upper_estimate": 0.003788613808014744
+      "required_token_positions_per_second": 9299.445181305227,
+      "required_pages_per_second_upper_estimate": 0.0037096877219184726
     },
     {
       "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": 18994.59418786272,
-      "required_pages_per_second_upper_estimate": 0.004394065071921873
+      "required_token_positions_per_second": 18598.890362610455,
+      "required_pages_per_second_upper_estimate": 0.00430252595609923
     },
     {
       "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": 28491.89128179408,
-      "required_pages_per_second_upper_estimate": 0.0017350066936605445
+      "required_token_positions_per_second": 27898.33554391568,
+      "required_pages_per_second_upper_estimate": 0.0016988622633700326
     },
     {
       "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": 28491.89128179408,
-      "required_pages_per_second_upper_estimate": 0.0021933846614755194
+      "required_token_positions_per_second": 27898.33554391568,
+      "required_pages_per_second_upper_estimate": 0.0021476910977061966
     },
     {
       "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": 28491.89128179408,
-      "required_pages_per_second_upper_estimate": 0.35377393349384856
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27898.33554391568,
+      "required_pages_per_second_upper_estimate": 0.34640395773266547
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.10657264695311142,
-    "producer_p95_seconds": 0.20790532299997722,
-    "gpu_update_seconds": 0.9703813525944024,
-    "producer_batches_per_second": 9.383270741506198,
-    "gpu_batches_per_second": 1.0305226881435938,
-    "producer_headroom": 9.105350953902265,
+    "producer_batch_seconds": 0.05207734765628658,
+    "producer_p95_seconds": 0.1905860759998177,
+    "gpu_update_seconds": 0.9910268645410183,
+    "producer_batches_per_second": 19.20220681360457,
+    "gpu_batches_per_second": 1.0090543816520428,
+    "producer_headroom": 19.029902810754713,
     "sustainable": true,
-    "buffer_seconds": 60.16364386085295,
+    "buffer_seconds": 60.45263873700212,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 464565.3013896842,
+    "generation_bytes_per_second": 465279.3152191602,
     "upload_bytes_per_second": 5695084.725803528,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.68489915149588,
+    "estimated_full_shards_per_hour": 99.83810989791017,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 362100, "loss": 0.5817650128155947, "lr": 0.0003, "grad_norm": 0.1426328456273929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53001223874, "tokens_per_second": 147798.26814777747, "data_wait_s": 1.5081000128702726e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 362200, "loss": 0.5812425687909126, "lr": 0.0003, "grad_norm": 0.14682848256624062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53015925490, "tokens_per_second": 150572.19699902638, "data_wait_s": 1.411500034009805e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 362300, "loss": 0.581356992945075, "lr": 0.0003, "grad_norm": 0.1601125700377772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53030612571, "tokens_per_second": 149150.71362877195, "data_wait_s": 7.561000529676676e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 362400, "loss": 0.6729161273688078, "lr": 0.0003, "grad_norm": 0.18195817486300464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53045348843, "tokens_per_second": 150079.46739753152, "data_wait_s": 8.346999493369367e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:881ff5b17fb1c69543d1969830c925d877ac16713a678d91163cdd4336caef16
-size 155562
+oid sha256:2c4510e83bdbc96f9962826b6699a6c2fc11140bdbf2421d443bf74d6e9085f3
+size 157680