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": 76217.88870531849,
-      "required_pages_per_second_upper_estimate": 0.005272930390500886
+      "required_token_positions_per_second": 76800.71477432967,
+      "required_pages_per_second_upper_estimate": 0.005313251650297876
     },
     {
       "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": 9527.236088164811,
-      "required_pages_per_second_upper_estimate": 0.003988854798801578
+      "required_token_positions_per_second": 9600.089346791208,
+      "required_pages_per_second_upper_estimate": 0.004019356936839412
     },
     {
       "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": 19054.472176329622,
-      "required_pages_per_second_upper_estimate": 0.003968544054106494
+      "required_token_positions_per_second": 19200.178693582417,
+      "required_pages_per_second_upper_estimate": 0.003998890879110992
     },
     {
       "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": 28581.70826449443,
-      "required_pages_per_second_upper_estimate": 0.00171342976871486
+      "required_token_positions_per_second": 28800.268040373627,
+      "required_pages_per_second_upper_estimate": 0.0017265320935573633
     },
     {
       "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": 28581.70826449443,
-      "required_pages_per_second_upper_estimate": 0.0022221020999160374
+      "required_token_positions_per_second": 28800.268040373627,
+      "required_pages_per_second_upper_estimate": 0.0022390941611477883
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28581.70826449443,
-      "required_pages_per_second_upper_estimate": 0.3666154649695929
+      "required_token_positions_per_second": 28800.268040373627,
+      "required_pages_per_second_upper_estimate": 0.3694189151033674
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09568031653122944,
-    "producer_p95_seconds": 0.1666256489997977,
-    "gpu_update_seconds": 0.9673319643509787,
-    "producer_batches_per_second": 10.451470440878051,
-    "gpu_batches_per_second": 1.0337712769276053,
-    "producer_headroom": 10.110041431930755,
+    "producer_batch_seconds": 0.04947380697656456,
+    "producer_p95_seconds": 0.16945143100019777,
+    "gpu_update_seconds": 0.959991065404033,
+    "producer_batches_per_second": 20.212715800781087,
+    "gpu_batches_per_second": 1.0416763614139766,
+    "producer_headroom": 19.404026576300765,
     "sustainable": true,
-    "buffer_seconds": 60.94191375411166,
+    "buffer_seconds": 60.47943712045408,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 469773.5105310691,
-    "upload_bytes_per_second": 4316770.200713523,
+    "generation_bytes_per_second": 469559.4034046904,
+    "upload_bytes_per_second": 4361792.550967564,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.80245959233336,
+    "estimated_full_shards_per_hour": 100.7565171871713,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 7030636,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.0609669329996905,
+      "bytes": 7430334,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 847500, "loss": 0.5078483819961548, "lr": 3e-05, "grad_norm": 0.29790460234382193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124050872630, "tokens_per_second": 151769.51493085473, "data_wait_s": 9.835000128077809e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 847600, "loss": 0.5515899658203125, "lr": 3e-05, "grad_norm": 0.2993019438846067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124065418168, "tokens_per_second": 151187.85278245917, "data_wait_s": 1.8619000002217945e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 847700, "loss": 0.5003780126571655, "lr": 3e-05, "grad_norm": 0.29074722812008036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124079941444, "tokens_per_second": 151178.24556929132, "data_wait_s": 1.4152000403555576e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 847800, "loss": 0.49885714054107666, "lr": 3e-05, "grad_norm": 0.2845417904857321, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124094580343, "tokens_per_second": 152317.1407648859, "data_wait_s": 9.455000508751255e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71fdb76fed2ac91dc25b9e94e1e0f940655de10d1583a5677b709f7a2b84b675
-size 111084
+oid sha256:19a42f9f094a5f40dc93f984c06609c32d6ed60c74b2c9022ea2a80205ea2697
+size 113202