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": 75583.06335444357,
-      "required_pages_per_second_upper_estimate": 0.006551348368040456
+      "required_token_positions_per_second": 74992.82310360523,
+      "required_pages_per_second_upper_estimate": 0.006500187839048027
     },
     {
       "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": 9447.882919305446,
-      "required_pages_per_second_upper_estimate": 0.003768901754948718
+      "required_token_positions_per_second": 9374.102887950654,
+      "required_pages_per_second_upper_estimate": 0.00373946979733152
     },
     {
       "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": 18895.765838610892,
-      "required_pages_per_second_upper_estimate": 0.004371202872642004
+      "required_token_positions_per_second": 18748.205775901308,
+      "required_pages_per_second_upper_estimate": 0.004337067449102548
     },
     {
       "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": 28343.648757916337,
-      "required_pages_per_second_upper_estimate": 0.0017259795017248035
+      "required_token_positions_per_second": 28122.30866385196,
+      "required_pages_per_second_upper_estimate": 0.0017125010512780078
     },
     {
       "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": 28343.648757916337,
-      "required_pages_per_second_upper_estimate": 0.0021819725416258404
+      "required_token_positions_per_second": 28122.30866385196,
+      "required_pages_per_second_upper_estimate": 0.002164933168476169
     },
     {
       "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": 28343.648757916337,
-      "required_pages_per_second_upper_estimate": 0.3519332574830989
+      "required_token_positions_per_second": 28122.30866385196,
+      "required_pages_per_second_upper_estimate": 0.3491849542924613
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11979209478852226,
-    "producer_p95_seconds": 0.38984344500204315,
-    "gpu_update_seconds": 0.975456626496543,
-    "producer_batches_per_second": 8.34779625287773,
-    "gpu_batches_per_second": 1.0251609070426915,
-    "producer_headroom": 8.142913171512593,
+    "producer_batch_seconds": 0.09806024578159622,
+    "producer_p95_seconds": 0.39671837799687637,
+    "gpu_update_seconds": 0.9831340780189348,
+    "producer_batches_per_second": 10.197812498117134,
+    "gpu_batches_per_second": 1.0171552612793677,
+    "producer_headroom": 10.02581698814636,
     "sustainable": true,
-    "buffer_seconds": 60.47831084278566,
+    "buffer_seconds": 60.95431283717396,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379734.8500496105,
-    "upload_bytes_per_second": 4092962.040476732,
+    "generation_bytes_per_second": 380111.8026197752,
+    "upload_bytes_per_second": 4014829.5622885916,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.48225904575573,
+    "estimated_full_shards_per_hour": 81.56314429230635,
     "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": 1.1287752040007035,
-      "bytes": 10564318,
-      "pages": 35,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0117927669998608,
-      "bytes": 847459,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.3560152030004247,
       "bytes": 13557643,
@@ -562,6 +550,18 @@
       "bytes": 13413721,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.8657637560027069,
+      "bytes": 989817,
+      "pages": 27,
+      "shards": 1
+    },
+    {
+      "seconds": 0.8787106930030859,
+      "bytes": 615085,
+      "pages": 17,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -348,3 +348,4 @@
 {"step": 349800, "loss": 0.5988677442073822, "lr": 0.0003, "grad_norm": 0.17552263591675815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51201641908, "tokens_per_second": 148145.49559771136, "data_wait_s": 1.6423997294623405e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 349900, "loss": 0.584590258076787, "lr": 0.0003, "grad_norm": 0.17714293403909717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51216225970, "tokens_per_second": 147863.93379601528, "data_wait_s": 9.56900476012379e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 350000, "loss": 0.6504786228761077, "lr": 0.0003, "grad_norm": 0.1511979477375556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51230919415, "tokens_per_second": 148850.98533615225, "data_wait_s": 1.4363999071065336e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 350100, "loss": 0.693084504455328, "lr": 0.0003, "grad_norm": 0.18748733395407843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51245680972, "tokens_per_second": 135294.73832244033, "data_wait_s": 1.1290001566521823e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b68b38b5d61dbac65cdd4618e48c1db1ccbc57d75138fd561f4ee3189ee29692
-size 746946
+oid sha256:72f30e1aa68e9e7d16e526cfa697f4a6eb8dfdb507f418b8fc791317e1170e8b
+size 749064