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": 76209.47403154314,
-      "required_pages_per_second_upper_estimate": 0.0066056440579081
+      "required_token_positions_per_second": 76073.89670516354,
+      "required_pages_per_second_upper_estimate": 0.006593892558875109
     },
     {
       "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": 9526.184253942893,
-      "required_pages_per_second_upper_estimate": 0.003800137328044875
+      "required_token_positions_per_second": 9509.237088145443,
+      "required_pages_per_second_upper_estimate": 0.0037933768502255633
     },
     {
       "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": 19052.368507885785,
-      "required_pages_per_second_upper_estimate": 0.004407430144065904
+      "required_token_positions_per_second": 19018.474176290885,
+      "required_pages_per_second_upper_estimate": 0.004399589286971287
     },
     {
       "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": 28578.552761828676,
-      "required_pages_per_second_upper_estimate": 0.001740283923117005
+      "required_token_positions_per_second": 28527.71126443633,
+      "required_pages_per_second_upper_estimate": 0.0017371879426705326
     },
     {
       "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": 28578.552761828676,
-      "required_pages_per_second_upper_estimate": 0.0022000561021029135
+      "required_token_positions_per_second": 28527.71126443633,
+      "required_pages_per_second_upper_estimate": 0.002196142182895378
     },
     {
       "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": 28578.552761828676,
-      "required_pages_per_second_upper_estimate": 0.3548499790385621
+      "required_token_positions_per_second": 28527.71126443633,
+      "required_pages_per_second_upper_estimate": 0.3542186977964952
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.15464321065616105,
-    "producer_p95_seconds": 0.37222080300125526,
-    "gpu_update_seconds": 0.9674387723694818,
-    "producer_batches_per_second": 6.466497919675464,
-    "gpu_batches_per_second": 1.033657145610123,
-    "producer_headroom": 6.25594080894064,
+    "producer_batch_seconds": 0.10227536282792471,
+    "producer_p95_seconds": 0.3752321319989278,
+    "gpu_update_seconds": 0.9691629217541539,
+    "producer_batches_per_second": 9.777525812178936,
+    "gpu_batches_per_second": 1.0318182604324482,
+    "producer_headroom": 9.476015483657994,
     "sustainable": true,
-    "buffer_seconds": 60.948642659277354,
+    "buffer_seconds": 60.08810114875754,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374016.0725113873,
-    "upload_bytes_per_second": 4945261.915055714,
+    "generation_bytes_per_second": 392941.70644077903,
+    "upload_bytes_per_second": 4937298.501192445,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.25514251238073,
+    "estimated_full_shards_per_hour": 84.31614298741844,
     "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": 1.2891920199981541,
-      "bytes": 7891753,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 0.9012639180000406,
       "bytes": 5714275,
@@ -562,6 +556,12 @@
       "bytes": 2398655,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0890797859974555,
+      "bytes": 6225655,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -383,3 +383,4 @@
 {"step": 193300, "loss": 0.6072239205241203, "lr": 0.0003, "grad_norm": 0.16099361596421796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28291318837, "tokens_per_second": 150242.16291948615, "data_wait_s": 1.2869000784121454e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 193400, "loss": 0.5866144802421331, "lr": 0.0003, "grad_norm": 0.14337903761812446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28305891675, "tokens_per_second": 149803.2884866767, "data_wait_s": 9.187999239657074e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 193500, "loss": 0.6188144702464342, "lr": 0.0003, "grad_norm": 0.15031023066255766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28320553337, "tokens_per_second": 150369.8714985932, "data_wait_s": 8.048999006859958e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 193600, "loss": 0.5718002542853355, "lr": 0.0003, "grad_norm": 0.14237183209317347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28335333420, "tokens_per_second": 151310.95929628066, "data_wait_s": 1.0493000445421785e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dce830e6db19baec1eeb966df0bc1ae3653c48686eefec78bea4975231469649
-size 821076
+oid sha256:3a236697ed3a0305570aac3a391ba09c689a6b812849d410deea7681f7793183
+size 823194