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": 73299.30631199596,
-      "required_pages_per_second_upper_estimate": 0.0063533980957304025
+      "required_token_positions_per_second": 74255.28093640812,
+      "required_pages_per_second_upper_estimate": 0.006436259553278915
     },
     {
       "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": 9162.413288999494,
-      "required_pages_per_second_upper_estimate": 0.0036550236512683477
+      "required_token_positions_per_second": 9281.910117051015,
+      "required_pages_per_second_upper_estimate": 0.0037026927226148933
     },
     {
       "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": 18324.82657799899,
-      "required_pages_per_second_upper_estimate": 0.004239126122887239
+      "required_token_positions_per_second": 18563.82023410203,
+      "required_pages_per_second_upper_estimate": 0.004294413099081987
     },
     {
       "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": 27487.239866998483,
-      "required_pages_per_second_upper_estimate": 0.0016738286934981016
+      "required_token_positions_per_second": 27845.730351153044,
+      "required_pages_per_second_upper_estimate": 0.0016956588831289043
     },
     {
       "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": 27487.239866998483,
-      "required_pages_per_second_upper_estimate": 0.002116043814511443
+      "required_token_positions_per_second": 27845.730351153044,
+      "required_pages_per_second_upper_estimate": 0.0021436414043468355
     },
     {
       "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": 27487.239866998483,
-      "required_pages_per_second_upper_estimate": 0.3412995252740788
+      "required_token_positions_per_second": 27845.730351153044,
+      "required_pages_per_second_upper_estimate": 0.34575077729680825
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.13219398896092116,
-    "producer_p95_seconds": 0.45185798699822044,
-    "gpu_update_seconds": 1.0058485367675831,
-    "producer_batches_per_second": 7.564640479194688,
-    "gpu_batches_per_second": 0.9941854697265077,
-    "producer_headroom": 7.608882557170806,
+    "producer_batch_seconds": 0.1390393619765291,
+    "producer_p95_seconds": 0.4389112680000835,
+    "gpu_update_seconds": 0.992899078291015,
+    "producer_batches_per_second": 7.192207917127868,
+    "gpu_batches_per_second": 1.0071517054091814,
+    "producer_headroom": 7.1411366117936,
     "sustainable": true,
-    "buffer_seconds": 60.350912206054986,
+    "buffer_seconds": 60.56684377575191,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302594.4528625146,
+    "generation_bytes_per_second": 327433.3187705911,
     "upload_bytes_per_second": 5667045.8685436025,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.92972554594591,
+    "estimated_full_shards_per_hour": 70.25956795061397,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 236600, "loss": 0.5741153005510569, "lr": 0.0003, "grad_norm": 0.16989412711838428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34631263270, "tokens_per_second": 145692.82722657136, "data_wait_s": 1.1872005416080356e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 236700, "loss": 0.6095123514533043, "lr": 0.0003, "grad_norm": 0.14309083128939812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34645759926, "tokens_per_second": 146735.131508522, "data_wait_s": 1.2876997061539441e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 236800, "loss": 0.6036998704075813, "lr": 0.0003, "grad_norm": 0.15617763721920008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34660451462, "tokens_per_second": 148662.44309618772, "data_wait_s": 1.107400021282956e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 236900, "loss": 0.6146441139280796, "lr": 0.0003, "grad_norm": 0.1577055018214105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34674944830, "tokens_per_second": 147353.71542987463, "data_wait_s": 1.2205004168208688e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:486de5737aece2e697c4b09c4014e295d3babae0cc0f39370e42f05d23e2b0b5
-size 891753
+oid sha256:8446ab8354c739e1c01012b38500b1b105be3d7545ba968741b2bd23d4ad0975
+size 893871