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": 77226.44308856814,
-      "required_pages_per_second_upper_estimate": 0.005342704522902678
+      "required_token_positions_per_second": 77040.86677931425,
+      "required_pages_per_second_upper_estimate": 0.005329865923232133
     },
     {
       "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": 9653.305386071017,
-      "required_pages_per_second_upper_estimate": 0.0040416373812086
+      "required_token_positions_per_second": 9630.108347414282,
+      "required_pages_per_second_upper_estimate": 0.004031925265532275
     },
     {
       "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": 19306.610772142034,
-      "required_pages_per_second_upper_estimate": 0.004021057874272299
+      "required_token_positions_per_second": 19260.216694828563,
+      "required_pages_per_second_upper_estimate": 0.004011395211461998
     },
     {
       "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": 28959.91615821305,
-      "required_pages_per_second_upper_estimate": 0.0017361027544532755
+      "required_token_positions_per_second": 28890.325042242846,
+      "required_pages_per_second_upper_estimate": 0.0017319308733103431
     },
     {
       "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": 28959.91615821305,
-      "required_pages_per_second_upper_estimate": 0.002251506100091946
+      "required_token_positions_per_second": 28890.325042242846,
+      "required_pages_per_second_upper_estimate": 0.0022460956969242323
     },
     {
       "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": 28959.91615821305,
-      "required_pages_per_second_upper_estimate": 0.3714667097422179
+      "required_token_positions_per_second": 28890.325042242846,
+      "required_pages_per_second_upper_estimate": 0.37057406962767087
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07819835046106505,
-    "producer_p95_seconds": 0.20879705399966042,
-    "gpu_update_seconds": 0.9546988965352723,
-    "producer_batches_per_second": 12.787993533161545,
-    "gpu_batches_per_second": 1.047450671231664,
-    "producer_headroom": 12.208683315009525,
+    "producer_batch_seconds": 0.0777804393359105,
+    "producer_p95_seconds": 0.20948546899853682,
+    "gpu_update_seconds": 0.9569985785751339,
+    "producer_batches_per_second": 12.8567029003436,
+    "gpu_batches_per_second": 1.0449336314468622,
+    "producer_headroom": 12.303846400791628,
     "sustainable": true,
-    "buffer_seconds": 60.14603048172216,
+    "buffer_seconds": 60.29091045023344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400093.44022937445,
-    "upload_bytes_per_second": 4308850.019408859,
+    "generation_bytes_per_second": 415516.3384747967,
+    "upload_bytes_per_second": 4275807.893389077,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.85073857460904,
+    "estimated_full_shards_per_hour": 89.16013351138044,
     "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": [
-    {
-      "seconds": 1.1799431950003054,
-      "bytes": 8208196,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.0678595420013153,
       "bytes": 6041574,
@@ -562,6 +556,12 @@
       "bytes": 7591094,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.4399703239996597,
+      "bytes": 6008739,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -163,3 +163,4 @@
 {"step": 701300, "loss": 0.5354167222976685, "lr": 8e-05, "grad_norm": 0.21774947897899255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102650876683, "tokens_per_second": 151278.39324420216, "data_wait_s": 1.918900125019718e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 701400, "loss": 0.5629359483718872, "lr": 8e-05, "grad_norm": 0.21888205249817885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102665495894, "tokens_per_second": 152786.00998708245, "data_wait_s": 1.655999949434772e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 701500, "loss": 0.5888649225234985, "lr": 8e-05, "grad_norm": 0.23307399384670752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102680184060, "tokens_per_second": 152914.71957144633, "data_wait_s": 1.8877999536925927e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 701600, "loss": 0.6164185404777527, "lr": 8e-05, "grad_norm": 0.22690946397067135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102694853102, "tokens_per_second": 153106.78212982367, "data_wait_s": 1.749800139805302e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:966141e4b68065475fa580180b4721ec3d0e0ebfe578a1259a41cb44cff558b7
-size 351984
+oid sha256:93588edff36a919e1d79894e68c1618c5f30848d08ab50757a664c60b8be96b9
+size 354102