Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75874.55349007812,
-      "required_pages_per_second_upper_estimate": 0.005249177663663102
+      "required_token_positions_per_second": 76648.2171317306,
+      "required_pages_per_second_upper_estimate": 0.005302701509539592
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9484.319186259765,
-      "required_pages_per_second_upper_estimate": 0.003970886388180757
+      "required_token_positions_per_second": 9581.027141466326,
+      "required_pages_per_second_upper_estimate": 0.004011375989533996
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18968.63837251953,
-      "required_pages_per_second_upper_estimate": 0.003950667136362478
+      "required_token_positions_per_second": 19162.05428293265,
+      "required_pages_per_second_upper_estimate": 0.0039909505697809655
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28452.957558779297,
-      "required_pages_per_second_upper_estimate": 0.0017057113604981825
+      "required_token_positions_per_second": 28743.081424398977,
+      "required_pages_per_second_upper_estimate": 0.0017231038432485896
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28452.957558779297,
-      "required_pages_per_second_upper_estimate": 0.00221209229886119
+      "required_token_positions_per_second": 28743.081424398977,
+      "required_pages_per_second_upper_estimate": 0.0022346481532930997
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28452.957558779297,
-      "required_pages_per_second_upper_estimate": 0.3649639891584163
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28743.081424398977,
+      "required_pages_per_second_upper_estimate": 0.36868538659584893
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.20132111199984593,
-    "producer_p95_seconds": 0.2761918809992494,
-    "gpu_update_seconds": 0.9717091779609771,
-    "producer_batches_per_second": 4.967188935459314,
-    "gpu_batches_per_second": 1.029114495036867,
-    "producer_headroom": 4.826663077252031,
+    "producer_batch_seconds": 0.39169291549211493,
+    "producer_p95_seconds": 0.321863837998535,
+    "gpu_update_seconds": 0.9619010429595276,
+    "producer_batches_per_second": 2.5530203903321063,
+    "gpu_batches_per_second": 1.039607979759801,
+    "producer_headroom": 2.4557529761573935,
     "sustainable": true,
-    "buffer_seconds": 60.24596903358058,
+    "buffer_seconds": 60.59976570645024,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 481701.8426173263,
+    "generation_bytes_per_second": 431250.9707560099,
     "upload_bytes_per_second": 5331039.643899832,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.36200198068468,
+    "estimated_full_shards_per_hour": 92.53641931543564,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -282,3 +282,4 @@
 {"step": 632600, "loss": 0.5609431862831116, "lr": 0.0001, "grad_norm": 0.17249366517498674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92595899835, "tokens_per_second": 151223.85709951268, "data_wait_s": 1.7721002222970128e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 632700, "loss": 0.5434319972991943, "lr": 0.0001, "grad_norm": 0.17139880844568653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92610515706, "tokens_per_second": 150756.4351332831, "data_wait_s": 1.1137999536003917e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 632800, "loss": 0.5363110303878784, "lr": 0.0001, "grad_norm": 0.17411284377927336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92625063190, "tokens_per_second": 150362.43460635113, "data_wait_s": 8.797000191407278e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 632900, "loss": 0.5373297333717346, "lr": 0.0001, "grad_norm": 0.15873660495925473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92639607102, "tokens_per_second": 151211.68209369996, "data_wait_s": 1.5767000149935484e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ba6b75af53cd077f18e0dcfaf8a88142cc17f9a7775d2f2ad0d191eb885a854a
-size 606375
+oid sha256:0bf273c1941f3be3d05d4a77c53cba61faae2c33d765dede84b75d95112ddbe2
+size 608493