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": 75045.71467202636,
-      "required_pages_per_second_upper_estimate": 0.006504772346146316
+      "required_token_positions_per_second": 75688.01669368563,
+      "required_pages_per_second_upper_estimate": 0.006560445457484151
     },
     {
       "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": 9380.714334003294,
-      "required_pages_per_second_upper_estimate": 0.0037421072020118453
+      "required_token_positions_per_second": 9461.002086710703,
+      "required_pages_per_second_upper_estimate": 0.0037741351869757076
     },
     {
       "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": 18761.42866800659,
-      "required_pages_per_second_upper_estimate": 0.004340126332476151
+      "required_token_positions_per_second": 18922.004173421406,
+      "required_pages_per_second_upper_estimate": 0.004377272649621495
     },
     {
       "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": 28142.14300200988,
-      "required_pages_per_second_upper_estimate": 0.0017137088584090955
+      "required_token_positions_per_second": 28383.00626013211,
+      "required_pages_per_second_upper_estimate": 0.0017283761671168887
     },
     {
       "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": 28142.14300200988,
-      "required_pages_per_second_upper_estimate": 0.002166460070732528
+      "required_token_positions_per_second": 28383.00626013211,
+      "required_pages_per_second_upper_estimate": 0.00218500239109495
     },
     {
       "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": 28142.14300200988,
-      "required_pages_per_second_upper_estimate": 0.3494312303911231
+      "required_token_positions_per_second": 28383.00626013211,
+      "required_pages_per_second_upper_estimate": 0.3524219459395323
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08474027024200836,
-    "producer_p95_seconds": 0.2214078069991956,
-    "gpu_update_seconds": 0.982441173652817,
-    "producer_batches_per_second": 11.800764821071684,
-    "gpu_batches_per_second": 1.0178726490888992,
-    "producer_headroom": 11.59355724081454,
+    "producer_batch_seconds": 0.42772661279681756,
+    "producer_p95_seconds": 0.2891327180004737,
+    "gpu_update_seconds": 0.9741040024655694,
+    "producer_batches_per_second": 2.3379419705994042,
+    "gpu_batches_per_second": 1.0265844278114913,
+    "producer_headroom": 2.2773986310931202,
     "sustainable": true,
-    "buffer_seconds": 60.91135276647466,
+    "buffer_seconds": 60.3944481528653,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306532.91108998744,
-    "upload_bytes_per_second": 5286393.6349620335,
+    "generation_bytes_per_second": 277390.4928969109,
+    "upload_bytes_per_second": 5216832.307734021,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.77482699894635,
+    "estimated_full_shards_per_hour": 59.5215424554872,
     "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": 1.150664095999673,
-      "bytes": 11982825,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 0.9085387359991728,
       "bytes": 1084848,
@@ -562,6 +556,12 @@
       "bytes": 5686805,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.051009315000556,
+      "bytes": 5972773,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -193,3 +193,4 @@
 {"step": 256400, "loss": 0.6124392151832581, "lr": 0.0003, "grad_norm": 0.1545684037659675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37530010083, "tokens_per_second": 147835.9191612433, "data_wait_s": 1.2881999282399192e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256500, "loss": 0.5793410521000624, "lr": 0.0003, "grad_norm": 0.14140056791090877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37544598640, "tokens_per_second": 149039.37765120564, "data_wait_s": 1.3975000911159441e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256600, "loss": 0.5796477869153023, "lr": 0.0003, "grad_norm": 0.15295621146090357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37559203520, "tokens_per_second": 148569.259743464, "data_wait_s": 1.5387999155791476e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 256700, "loss": 0.5793802235275507, "lr": 0.0003, "grad_norm": 0.1362063822111426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37573826960, "tokens_per_second": 149937.26378596557, "data_wait_s": 1.3355998817132786e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f8d875321d9862f0715ea5c1234ddf2951ad22dfe823044fb541fed2b4ed5c85
-size 416307
+oid sha256:95d8c5d8fa2d2d3595c6b409ac18bc0010987b3e6841aa823e7775757015f0ff
+size 418425