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": 76764.09997329117,
-      "required_pages_per_second_upper_estimate": 0.00531071855340399
+      "required_token_positions_per_second": 75240.38214393613,
+      "required_pages_per_second_upper_estimate": 0.005205304218456746
     },
     {
       "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": 9595.512496661397,
-      "required_pages_per_second_upper_estimate": 0.004017440705265558
+      "required_token_positions_per_second": 9405.047767992017,
+      "required_pages_per_second_upper_estimate": 0.003937697100727504
     },
     {
       "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": 19191.024993322793,
-      "required_pages_per_second_upper_estimate": 0.003996984404746221
+      "required_token_positions_per_second": 18810.095535984034,
+      "required_pages_per_second_upper_estimate": 0.003917646844567898
     },
     {
       "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": 28786.537489984188,
-      "required_pages_per_second_upper_estimate": 0.0017257089680268527
+      "required_token_positions_per_second": 28215.14330397605,
+      "required_pages_per_second_upper_estimate": 0.0016914547590440694
     },
     {
       "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": 28786.537489984188,
-      "required_pages_per_second_upper_estimate": 0.002238026671249318
+      "required_token_positions_per_second": 28215.14330397605,
+      "required_pages_per_second_upper_estimate": 0.002193603286584596
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28786.537489984188,
-      "required_pages_per_second_upper_estimate": 0.3692427943456881
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28215.14330397605,
+      "required_pages_per_second_upper_estimate": 0.36191356324285284
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.044507669203127875,
-    "producer_p95_seconds": 0.14738937200036162,
-    "gpu_update_seconds": 0.9604489601995264,
-    "producer_batches_per_second": 22.468037933779797,
-    "gpu_batches_per_second": 1.0411797413912105,
-    "producer_headroom": 21.579403671222323,
+    "producer_batch_seconds": 0.06728741275000871,
+    "producer_p95_seconds": 0.15887726099981592,
+    "gpu_update_seconds": 0.979899329311713,
+    "producer_batches_per_second": 14.861620608229888,
+    "gpu_batches_per_second": 1.020512995658856,
+    "producer_headroom": 14.5628920664896,
     "sustainable": true,
-    "buffer_seconds": 60.50828449257016,
+    "buffer_seconds": 60.7537584173262,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289702.9369927744,
-    "upload_bytes_per_second": 5452255.476764406,
+    "generation_bytes_per_second": 289978.5163001468,
+    "upload_bytes_per_second": 5313072.95235923,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.16350633943009,
+    "estimated_full_shards_per_hour": 62.222639243634255,
     "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": [
@@ -322,6 +322,18 @@
       "bytes": 11675037,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 0.8115306640002018,
+      "bytes": 1206731,
+      "pages": 34,
+      "shards": 1
+    },
+    {
+      "seconds": 1.1671833450000122,
+      "bytes": 5654442,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -27,3 +27,4 @@
 {"step": 399800, "loss": 0.5576823353767395, "lr": 0.0003, "grad_norm": 0.1638329301772677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58519792173, "tokens_per_second": 131438.76742203868, "data_wait_s": 1.1736000033124583e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 399900, "loss": 0.5426488518714905, "lr": 0.0003, "grad_norm": 0.14544560852037164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58534476130, "tokens_per_second": 152363.50477622534, "data_wait_s": 9.141000191448256e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 400000, "loss": 0.6140549182891846, "lr": 0.0003, "grad_norm": 0.1563160900023543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58549240171, "tokens_per_second": 153493.45537227663, "data_wait_s": 1.6833999779919395e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 400100, "loss": 0.5728186964988708, "lr": 0.0003, "grad_norm": 0.16538521601705272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58563914364, "tokens_per_second": 135370.72874448416, "data_wait_s": 9.292999948229408e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5c03dbdd602690495cd5f098e2ba55dad0007112964e36cad71761a79ae5969
-size 62370
+oid sha256:c61f8e2182fec3ef6cdbb3141ba7f38043a2774d0c16f710c0f0b611f4d55993
+size 64488