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": 76910.94365108047,
-      "required_pages_per_second_upper_estimate": 0.005320877539757728
+      "required_token_positions_per_second": 76856.19531265003,
+      "required_pages_per_second_upper_estimate": 0.005317089922671463
     },
     {
       "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": 9613.867956385058,
-      "required_pages_per_second_upper_estimate": 0.00402512575294627
+      "required_token_positions_per_second": 9607.024414081254,
+      "required_pages_per_second_upper_estimate": 0.0040222605046930805
     },
     {
       "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": 19227.735912770117,
-      "required_pages_per_second_upper_estimate": 0.004004630321135049
+      "required_token_positions_per_second": 19214.048828162508,
+      "required_pages_per_second_upper_estimate": 0.004001779662363978
     },
     {
       "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": 28841.603869155173,
-      "required_pages_per_second_upper_estimate": 0.001729010113376661
+      "required_token_positions_per_second": 28821.073242243765,
+      "required_pages_per_second_upper_estimate": 0.0017277793336417742
     },
     {
       "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": 28841.603869155173,
-      "required_pages_per_second_upper_estimate": 0.002242307840018458
+      "required_token_positions_per_second": 28821.073242243765,
+      "required_pages_per_second_upper_estimate": 0.0022407116740807797
     },
     {
       "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": 28841.603869155173,
-      "required_pages_per_second_upper_estimate": 0.36994912673202207
+      "required_token_positions_per_second": 28821.073242243765,
+      "required_pages_per_second_upper_estimate": 0.36968578189407225
     }
   ],
   "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": 0.15141765675031138,
-    "producer_p95_seconds": 0.3586812439971254,
-    "gpu_update_seconds": 0.9586152048072583,
-    "producer_batches_per_second": 6.604249606431342,
-    "gpu_batches_per_second": 1.0431714362396982,
-    "producer_headroom": 6.330934089067436,
+    "producer_batch_seconds": 0.11193002053130385,
+    "producer_p95_seconds": 0.3596409120000317,
+    "gpu_update_seconds": 0.9592980721993253,
+    "producer_batches_per_second": 8.934153636828169,
+    "gpu_batches_per_second": 1.042428864375136,
+    "producer_headroom": 8.570516360541854,
     "sustainable": true,
-    "buffer_seconds": 60.39275790285727,
+    "buffer_seconds": 60.435778548557494,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355466.72319649643,
+    "generation_bytes_per_second": 341288.6033649705,
     "upload_bytes_per_second": 4850426.610615096,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.2748839561574,
+    "estimated_full_shards_per_hour": 73.2325894900497,
     "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

@@ -368,3 +368,4 @@
 {"step": 641200, "loss": 0.4942816495895386, "lr": 0.0001, "grad_norm": 0.18223169601382702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93855081379, "tokens_per_second": 152525.69629604687, "data_wait_s": 1.374800194753334e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 641300, "loss": 0.5913184881210327, "lr": 0.0001, "grad_norm": 0.1737603321876027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93869753926, "tokens_per_second": 152536.7912513105, "data_wait_s": 1.5837998944334686e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 641400, "loss": 0.6155778765678406, "lr": 0.0001, "grad_norm": 0.19187098915342324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93884457422, "tokens_per_second": 151538.44337456545, "data_wait_s": 9.843999578151852e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 641500, "loss": 0.5121535062789917, "lr": 0.0001, "grad_norm": 0.1780030994972999, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93899177171, "tokens_per_second": 152155.189234055, "data_wait_s": 1.7579004634171724e-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 sha256:80cf886c984ccf84176ad7e7bc249d16fddd2b15c24c459e7dc7fa1de3bed610
-size 790089
+oid sha256:a6a9c613f9ff2b47e4a6894bb2970fe3d27445ab91906bd4f7f95310c0678155
+size 792207