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": 76169.05853121693,
-      "required_pages_per_second_upper_estimate": 0.006602140944771973
+      "required_token_positions_per_second": 75206.02738299112,
+      "required_pages_per_second_upper_estimate": 0.00651866784562389
     },
     {
       "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": 9521.132316402116,
-      "required_pages_per_second_upper_estimate": 0.003798122034626662
+      "required_token_positions_per_second": 9400.75342287389,
+      "required_pages_per_second_upper_estimate": 0.0037501010941734044
     },
     {
       "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": 19042.264632804232,
-      "required_pages_per_second_upper_estimate": 0.004405092790387915
+      "required_token_positions_per_second": 18801.50684574778,
+      "required_pages_per_second_upper_estimate": 0.0043493977135447386
     },
     {
       "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": 28563.396949206348,
-      "required_pages_per_second_upper_estimate": 0.0017393610136447087
+      "required_token_positions_per_second": 28202.26026862167,
+      "required_pages_per_second_upper_estimate": 0.0017173696845348317
     },
     {
       "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": 28563.396949206348,
-      "required_pages_per_second_upper_estimate": 0.002198889365693272
+      "required_token_positions_per_second": 28202.26026862167,
+      "required_pages_per_second_upper_estimate": 0.0021710880643315486
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28563.396949206348,
-      "required_pages_per_second_upper_estimate": 0.3546617945690347
+      "required_token_positions_per_second": 28202.26026862167,
+      "required_pages_per_second_upper_estimate": 0.350177685642893
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08055057178124514,
-    "producer_p95_seconds": 0.21820405300059065,
-    "gpu_update_seconds": 0.9679520978952826,
-    "producer_batches_per_second": 12.414561161846816,
-    "gpu_batches_per_second": 1.033108975304049,
-    "producer_headroom": 12.016700521058922,
+    "producer_batch_seconds": 0.10113558959371005,
+    "producer_p95_seconds": 0.30211217899886833,
+    "gpu_update_seconds": 0.9803469557637424,
+    "producer_batches_per_second": 9.887716124633076,
+    "gpu_batches_per_second": 1.0200470293916981,
+    "producer_headroom": 9.693392402240105,
     "sustainable": true,
-    "buffer_seconds": 60.01303006950752,
+    "buffer_seconds": 60.78151125735203,
     "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": 330882.6605949245,
+    "generation_bytes_per_second": 319541.55449343595,
     "upload_bytes_per_second": 5933413.4313584715,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.99971640954783,
+    "estimated_full_shards_per_hour": 68.56617904760655,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -142,3 +142,4 @@
 {"step": 129200, "loss": 0.6225546598434448, "lr": 0.0003, "grad_norm": 0.1455784107006561, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18907853512, "tokens_per_second": 149300.0929345801, "data_wait_s": 9.02399915503338e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 129300, "loss": 0.6238899454474449, "lr": 0.0003, "grad_norm": 0.15132003403086974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18922486391, "tokens_per_second": 149880.6559437037, "data_wait_s": 1.3660999684361741e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 129400, "loss": 0.6272739339619875, "lr": 0.0003, "grad_norm": 0.13923481436885762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18937140674, "tokens_per_second": 148085.75099753367, "data_wait_s": 1.863300167315174e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 129500, "loss": 0.6373825017362833, "lr": 0.0003, "grad_norm": 0.13696854720360788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18951874620, "tokens_per_second": 150791.8481099114, "data_wait_s": 1.685600000200793e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b567edf53d606f2fd4f709a050646792827927128142f1a4e84112c408c14fe
-size 306723
+oid sha256:4c3fde1268e70f45f6cec59b42f9f3f75ffdfa86b4e8a79b06fc495711163f45
+size 308841