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": 76942.96702956906,
-      "required_pages_per_second_upper_estimate": 0.005323092991386044
+      "required_token_positions_per_second": 76293.94723189627,
+      "required_pages_per_second_upper_estimate": 0.005278192295849632
     },
     {
       "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": 9617.870878696132,
-      "required_pages_per_second_upper_estimate": 0.0040268016929273925
+      "required_token_positions_per_second": 9536.743403987033,
+      "required_pages_per_second_upper_estimate": 0.003992835313401011
     },
     {
       "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": 19235.741757392265,
-      "required_pages_per_second_upper_estimate": 0.004006297727441522
+      "required_token_positions_per_second": 19073.486807974066,
+      "required_pages_per_second_upper_estimate": 0.003972504300428475
     },
     {
       "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": 28853.612636088397,
-      "required_pages_per_second_upper_estimate": 0.0017297300206179822
+      "required_token_positions_per_second": 28610.2302119611,
+      "required_pages_per_second_upper_estimate": 0.0017151396159149949
     },
     {
       "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": 28853.612636088397,
-      "required_pages_per_second_upper_estimate": 0.0022432414688265327
+      "required_token_positions_per_second": 28610.2302119611,
+      "required_pages_per_second_upper_estimate": 0.0022243195558767823
     },
     {
       "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": 28853.612636088397,
-      "required_pages_per_second_upper_estimate": 0.37010316230023216
+      "required_token_positions_per_second": 28610.2302119611,
+      "required_pages_per_second_upper_estimate": 0.3669813138872141
     }
   ],
   "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": 0.18642643738270692,
-    "producer_p95_seconds": 0.23165708300075494,
-    "gpu_update_seconds": 0.9582162327021578,
-    "producer_batches_per_second": 5.364046076507607,
-    "gpu_batches_per_second": 1.0436057811085213,
-    "producer_headroom": 5.139916023471909,
+    "producer_batch_seconds": 0.23786467314059223,
+    "producer_p95_seconds": 0.22681780099992466,
+    "gpu_update_seconds": 0.9663676172882097,
+    "producer_batches_per_second": 4.204071108150391,
+    "gpu_batches_per_second": 1.034802886717343,
+    "producer_headroom": 4.0626781796934965,
     "sustainable": true,
-    "buffer_seconds": 60.367622660235945,
+    "buffer_seconds": 60.88115988915722,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 480694.27939603303,
-    "upload_bytes_per_second": 5198415.335503838,
+    "generation_bytes_per_second": 455441.6953324453,
+    "upload_bytes_per_second": 5077011.101608611,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.14580236826652,
+    "estimated_full_shards_per_hour": 97.7271856782915,
     "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": 1.2262302330000239,
-      "bytes": 12002858,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 0.976691177000248,
       "bytes": 6244802,
@@ -562,6 +556,12 @@
       "bytes": 8295904,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 0.88294980000137,
+      "bytes": 116127,
+      "pages": 4,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -154,3 +154,4 @@
 {"step": 619800, "loss": 0.5233141183853149, "lr": 0.0001, "grad_norm": 0.14545157516544222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90721640713, "tokens_per_second": 152442.2338918091, "data_wait_s": 1.400600012857467e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 619900, "loss": 0.5285357236862183, "lr": 0.0001, "grad_norm": 0.15875105633434458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90736332835, "tokens_per_second": 152907.69725344583, "data_wait_s": 1.3785998817184009e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 620000, "loss": 0.519646942615509, "lr": 0.0001, "grad_norm": 0.15932970628983065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90751225678, "tokens_per_second": 154578.89700448365, "data_wait_s": 9.052999303094111e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 620100, "loss": 0.6086562871932983, "lr": 0.0001, "grad_norm": 0.1629300007082568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90765923478, "tokens_per_second": 141751.49508206395, "data_wait_s": 1.1692000043694861e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:76e767da15cac5e022f103b9192174a6034a484b484d7c1e4e691a1f061fc257
-size 333705
+oid sha256:1fd8a78758ac99c9cc61f447375796136817aa965db7aa97b007d51fd82640b3
+size 335823