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": 75217.88994192265,
-      "required_pages_per_second_upper_estimate": 0.006519696062166695
+      "required_token_positions_per_second": 75836.16557119373,
+      "required_pages_per_second_upper_estimate": 0.006573286626706656
     },
     {
       "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": 9402.236242740331,
-      "required_pages_per_second_upper_estimate": 0.0037506926131882605
+      "required_token_positions_per_second": 9479.520696399217,
+      "required_pages_per_second_upper_estimate": 0.003781522537258344
     },
     {
       "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": 18804.472485480663,
-      "required_pages_per_second_upper_estimate": 0.004350083762103465
+      "required_token_positions_per_second": 18959.041392798434,
+      "required_pages_per_second_upper_estimate": 0.004385840558491573
     },
     {
       "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": 28206.708728220998,
-      "required_pages_per_second_upper_estimate": 0.001717640572385167
+      "required_token_positions_per_second": 28438.56208919765,
+      "required_pages_per_second_upper_estimate": 0.0017317592256280716
     },
     {
       "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": 28206.708728220998,
-      "required_pages_per_second_upper_estimate": 0.0021714305190656293
+      "required_token_positions_per_second": 28438.56208919765,
+      "required_pages_per_second_upper_estimate": 0.0021892792326047928
     },
     {
       "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": 28206.708728220998,
-      "required_pages_per_second_upper_estimate": 0.35023292062308004
+      "required_token_positions_per_second": 28438.56208919765,
+      "required_pages_per_second_upper_estimate": 0.35311176340312717
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.08728007167181318,
-    "producer_p95_seconds": 0.24834947300041676,
-    "gpu_update_seconds": 0.9801923459555562,
-    "producer_batches_per_second": 11.457369143326986,
-    "gpu_batches_per_second": 1.0202079256445673,
-    "producer_headroom": 11.230425539076478,
+    "producer_batch_seconds": 0.39397435940622927,
+    "producer_p95_seconds": 0.28140789399913047,
+    "gpu_update_seconds": 0.9722010526862065,
+    "producer_batches_per_second": 2.5382362484379195,
+    "gpu_batches_per_second": 1.0285938255641511,
+    "producer_headroom": 2.467675952697633,
     "sustainable": true,
-    "buffer_seconds": 60.771925449244485,
+    "buffer_seconds": 60.276465266544804,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 442261.68870318064,
-    "upload_bytes_per_second": 6356219.106286178,
+    "generation_bytes_per_second": 401176.1427255821,
+    "upload_bytes_per_second": 6217736.415972115,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.89906307050289,
+    "estimated_full_shards_per_hour": 86.08306132627104,
     "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": 1.181683423999857,
-      "bytes": 12141179,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.1883426349995716,
       "bytes": 1212298,
@@ -562,6 +556,12 @@
       "bytes": 6181281,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0534396790008032,
+      "bytes": 2069289,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -153,3 +153,4 @@
 {"step": 55300, "loss": 0.6829722598195076, "lr": 0.0003, "grad_norm": 0.13923664273611125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8093995532, "tokens_per_second": 150641.57423894596, "data_wait_s": 1.6832000255817547e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 55400, "loss": 0.6651799995452166, "lr": 0.0003, "grad_norm": 0.13794586564676423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8108646612, "tokens_per_second": 149414.72436736073, "data_wait_s": 1.3290000424603932e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 55500, "loss": 0.6675822697579861, "lr": 0.0003, "grad_norm": 0.13384163295611565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8123233060, "tokens_per_second": 150093.26572923284, "data_wait_s": 7.4460003816057e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 55600, "loss": 0.643940506502986, "lr": 0.0003, "grad_norm": 0.15360191239520263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8137899645, "tokens_per_second": 149318.89784144715, "data_wait_s": 1.585300014994573e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14ea9c9c03d125e8c99c21abc6efe0fbc5936af89afe7e825b962b770d4533cb
-size 330804
+oid sha256:96b8f023135f0c201304e0b02a945df652bbff03232648326b8322eeafead69a
+size 332922