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": 76198.92388452306,
-      "required_pages_per_second_upper_estimate": 0.005271618360193811
+      "required_token_positions_per_second": 76605.36130465951,
+      "required_pages_per_second_upper_estimate": 0.005299736643983595
     },
     {
       "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": 9524.865485565382,
-      "required_pages_per_second_upper_estimate": 0.003987862276997008
+      "required_token_positions_per_second": 9575.67016308244,
+      "required_pages_per_second_upper_estimate": 0.004009133134551092
     },
     {
       "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": 19049.730971130764,
-      "required_pages_per_second_upper_estimate": 0.003967556586097562
+      "required_token_positions_per_second": 19151.34032616488,
+      "required_pages_per_second_upper_estimate": 0.003988719135132279
     },
     {
       "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": 28574.596456696145,
-      "required_pages_per_second_upper_estimate": 0.0017130034266964545
+      "required_token_positions_per_second": 28727.010489247317,
+      "required_pages_per_second_upper_estimate": 0.0017221404152251447
     },
     {
       "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": 28574.596456696145,
-      "required_pages_per_second_upper_estimate": 0.002221549188141257
+      "required_token_positions_per_second": 28727.010489247317,
+      "required_pages_per_second_upper_estimate": 0.0022333987087736315
     },
     {
       "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": 28574.596456696145,
-      "required_pages_per_second_upper_estimate": 0.36652424233522074
+      "required_token_positions_per_second": 28727.010489247317,
+      "required_pages_per_second_upper_estimate": 0.3684792458953492
     }
   ],
   "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": 0.5074764133515828,
-    "producer_p95_seconds": 0.23750652099988656,
-    "gpu_update_seconds": 0.9675727194222893,
-    "producer_batches_per_second": 1.9705349326397046,
-    "gpu_batches_per_second": 1.0335140500830493,
-    "producer_headroom": 1.9066358434908166,
+    "producer_batch_seconds": 0.08125114501572739,
+    "producer_p95_seconds": 0.22050222300094902,
+    "gpu_update_seconds": 0.9624391654101565,
+    "producer_batches_per_second": 12.307518863967207,
+    "gpu_batches_per_second": 1.0390267104039106,
+    "producer_headroom": 11.845238183706357,
     "sustainable": true,
-    "buffer_seconds": 60.95708132360423,
+    "buffer_seconds": 60.633667420839856,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384331.0348991457,
-    "upload_bytes_per_second": 5311261.288170115,
+    "generation_bytes_per_second": 399855.59102233656,
+    "upload_bytes_per_second": 5156942.205517399,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.46849332075861,
+    "estimated_full_shards_per_hour": 85.79970167162487,
     "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": 1.115737056999933,
-      "bytes": 11496202,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.037322108000808,
       "bytes": 1047345,
@@ -562,6 +556,12 @@
       "bytes": 7299524,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 3.3975770770011877,
+      "bytes": 10836444,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -184,3 +184,4 @@
 {"step": 622800, "loss": 0.5662478804588318, "lr": 0.0001, "grad_norm": 0.16289424166216349, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91161265801, "tokens_per_second": 152186.00582515876, "data_wait_s": 1.0527000995352864e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 622900, "loss": 0.512753427028656, "lr": 0.0001, "grad_norm": 0.16105744409163986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91175779382, "tokens_per_second": 150315.8605156014, "data_wait_s": 1.1536998499650508e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 623000, "loss": 0.5415612459182739, "lr": 0.0001, "grad_norm": 0.15592117476941358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91190430891, "tokens_per_second": 151442.7507070528, "data_wait_s": 1.1983000149484724e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 623100, "loss": 0.557693362236023, "lr": 0.0001, "grad_norm": 0.15968168327072585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91205190000, "tokens_per_second": 153207.82871393726, "data_wait_s": 8.50900323712267e-06, "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 sha256:3c708a0671a3c432e8322e67d7b02fd919b8fbca35cce473a49c707683a9c3bb
-size 397245
+oid sha256:d1dc81f8f608091ff23c9f16468aa04a7228dbc3d7663b71a96ee5811808ae26
+size 399363