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": 75963.07544193706,
-      "required_pages_per_second_upper_estimate": 0.00525530181769193
+      "required_token_positions_per_second": 76817.15418253458,
+      "required_pages_per_second_upper_estimate": 0.005314388966702174
     },
     {
       "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": 9495.384430242133,
-      "required_pages_per_second_upper_estimate": 0.003975519174767605
+      "required_token_positions_per_second": 9602.144272816822,
+      "required_pages_per_second_upper_estimate": 0.004020217291454601
     },
     {
       "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": 18990.768860484266,
-      "required_pages_per_second_upper_estimate": 0.003955276333385318
+      "required_token_positions_per_second": 19204.288545633644,
+      "required_pages_per_second_upper_estimate": 0.003999746852909175
     },
     {
       "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": 28486.1532907264,
-      "required_pages_per_second_upper_estimate": 0.0017077013939414635
+      "required_token_positions_per_second": 28806.432818450467,
+      "required_pages_per_second_upper_estimate": 0.0017269016625899986
     },
     {
       "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": 28486.1532907264,
-      "required_pages_per_second_upper_estimate": 0.0022146731210074834
+      "required_token_positions_per_second": 28806.432818450467,
+      "required_pages_per_second_upper_estimate": 0.0022395734455272695
     },
     {
       "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": 28486.1532907264,
-      "required_pages_per_second_upper_estimate": 0.36538978836503383
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28806.432818450467,
+      "required_pages_per_second_upper_estimate": 0.3694979902573141
     }
   ],
   "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": 0.16520346511694584,
-    "producer_p95_seconds": 0.2830158489996393,
-    "gpu_update_seconds": 0.9705768173690458,
-    "producer_batches_per_second": 6.053141798763787,
-    "gpu_batches_per_second": 1.0303151508509258,
-    "producer_headroom": 5.875039102127697,
+    "producer_batch_seconds": 0.09978838639045762,
+    "producer_p95_seconds": 0.280802709999989,
+    "gpu_update_seconds": 0.9597856206024755,
+    "producer_batches_per_second": 10.021206236236186,
+    "gpu_batches_per_second": 1.0418993351580754,
+    "producer_headroom": 9.618209646631346,
     "sustainable": true,
-    "buffer_seconds": 60.17576267688084,
+    "buffer_seconds": 60.46649409795596,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448358.9971212266,
+    "generation_bytes_per_second": 447256.59954557585,
     "upload_bytes_per_second": 5172456.844281103,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.20740351893997,
+    "estimated_full_shards_per_hour": 95.97085466170746,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 750900, "loss": 0.5826548933982849, "lr": 5e-05, "grad_norm": 0.24639832661260694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109910790107, "tokens_per_second": 152329.28974206263, "data_wait_s": 1.2206997780594975e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 751000, "loss": 0.5846525430679321, "lr": 5e-05, "grad_norm": 0.26256670313217834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109925231035, "tokens_per_second": 148437.5651073467, "data_wait_s": 1.0769999789772555e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 751100, "loss": 0.4987703561782837, "lr": 5e-05, "grad_norm": 0.2295593934934252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109939798945, "tokens_per_second": 151038.21410001395, "data_wait_s": 1.4357003237819299e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 751200, "loss": 0.5086545944213867, "lr": 5e-05, "grad_norm": 0.22744074776930065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109954326672, "tokens_per_second": 149752.57790120298, "data_wait_s": 1.353399784420617e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb99109cc64d477ade5f8c4af7b5e78d6a2bb2524503975a554a246820481e11
-size 663561
+oid sha256:a68cf7c0ba1bc820a07fe499229b7f810add10f7671ed5e554bfecb89963d1f1
+size 665679