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": 76398.46932822168,
-      "required_pages_per_second_upper_estimate": 0.005285423376998101
+      "required_token_positions_per_second": 76439.97654272369,
+      "required_pages_per_second_upper_estimate": 0.005288294942407357
     },
     {
       "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": 9549.80866602771,
-      "required_pages_per_second_upper_estimate": 0.003998305465783746
+      "required_token_positions_per_second": 9554.99706784046,
+      "required_pages_per_second_upper_estimate": 0.0040004777412765945
     },
     {
       "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": 19099.61733205542,
-      "required_pages_per_second_upper_estimate": 0.003977946599486359
+      "required_token_positions_per_second": 19109.99413568092,
+      "required_pages_per_second_upper_estimate": 0.003980107814026837
     },
     {
       "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": 28649.425998083127,
-      "required_pages_per_second_upper_estimate": 0.0017174893434445102
+      "required_token_positions_per_second": 28664.991203521386,
+      "required_pages_per_second_upper_estimate": 0.0017184224537438412
     },
     {
       "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": 28649.425998083127,
-      "required_pages_per_second_upper_estimate": 0.0022273668558437268
+      "required_token_positions_per_second": 28664.991203521386,
+      "required_pages_per_second_upper_estimate": 0.0022285769821024483
     },
     {
       "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": 28649.425998083127,
-      "required_pages_per_second_upper_estimate": 0.36748407534643124
+      "required_token_positions_per_second": 28664.991203521386,
+      "required_pages_per_second_upper_estimate": 0.36768372908917774
     }
   ],
   "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": 0.079762360296872,
-    "producer_p95_seconds": 0.16723080399970058,
-    "gpu_update_seconds": 0.9650455126692545,
-    "producer_batches_per_second": 12.537241830332551,
-    "gpu_batches_per_second": 1.0362205583797428,
-    "producer_headroom": 12.0990089696117,
+    "producer_batch_seconds": 0.21979801755472295,
+    "producer_p95_seconds": 0.16460215300003256,
+    "gpu_update_seconds": 0.9645214890770156,
+    "producer_batches_per_second": 4.549631571408649,
+    "gpu_batches_per_second": 1.0367835360069946,
+    "producer_headroom": 4.388217418006873,
     "sustainable": true,
-    "buffer_seconds": 60.797867298163034,
+    "buffer_seconds": 60.764853811851985,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 469405.10464323446,
-    "upload_bytes_per_second": 3980294.179390897,
+    "generation_bytes_per_second": 447718.8218100861,
+    "upload_bytes_per_second": 4025407.9482820323,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.7234082648542,
+    "estimated_full_shards_per_hour": 96.07003679968774,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 5850227,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.452950692000286,
+      "bytes": 7702767,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -31,3 +31,4 @@
 {"step": 845600, "loss": 0.5235694050788879, "lr": 3e-05, "grad_norm": 0.2910009038958156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123772132582, "tokens_per_second": 152135.73965687054, "data_wait_s": 1.0953000128210988e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 845700, "loss": 0.5263145565986633, "lr": 3e-05, "grad_norm": 0.2982266387801221, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123786802260, "tokens_per_second": 152863.8770024828, "data_wait_s": 1.4893999832565896e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 845800, "loss": 0.5342761278152466, "lr": 3e-05, "grad_norm": 0.2988658951245049, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123801351839, "tokens_per_second": 150000.16667407998, "data_wait_s": 1.7228000160685042e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 845900, "loss": 0.5151355862617493, "lr": 3e-05, "grad_norm": 0.2853149347511223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123816069731, "tokens_per_second": 153360.16995992142, "data_wait_s": 9.447000138607109e-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 sha256:8679d3fc416ca9f170ba1121fc2cdeb1d78b1502df80a5c17b4896876a564916
-size 70842
+oid sha256:6e61b2f2550e2fe3fc0ac5cd39dc1814db3264207cf0220253ce29d4ef516b70
+size 72960