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": 76038.63139642138,
-      "required_pages_per_second_upper_estimate": 0.005260528953937122
+      "required_token_positions_per_second": 76360.80782159591,
+      "required_pages_per_second_upper_estimate": 0.005282817866582996
     },
     {
       "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": 9504.828924552672,
-      "required_pages_per_second_upper_estimate": 0.003979473387311959
+      "required_token_positions_per_second": 9545.100977699489,
+      "required_pages_per_second_upper_estimate": 0.003996334454988431
     },
     {
       "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": 19009.657849105344,
-      "required_pages_per_second_upper_estimate": 0.0039592104115789605
+      "required_token_positions_per_second": 19090.201955398978,
+      "required_pages_per_second_upper_estimate": 0.003975985624829002
     },
     {
       "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": 28514.486773658016,
-      "required_pages_per_second_upper_estimate": 0.0017093999429804905
+      "required_token_positions_per_second": 28635.30293309847,
+      "required_pages_per_second_upper_estimate": 0.0017166426872633515
     },
     {
       "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": 28514.486773658016,
-      "required_pages_per_second_upper_estimate": 0.002216875924679596
+      "required_token_positions_per_second": 28635.30293309847,
+      "required_pages_per_second_upper_estimate": 0.0022262688496619675
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28514.486773658016,
-      "required_pages_per_second_upper_estimate": 0.3657532198619568
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28635.30293309847,
+      "required_pages_per_second_upper_estimate": 0.36730291983297375
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1047224863279439,
-    "producer_p95_seconds": 0.2767974399976083,
-    "gpu_update_seconds": 0.9696124015650007,
-    "producer_batches_per_second": 9.54904753567871,
-    "gpu_batches_per_second": 1.0313399440703854,
-    "producer_headroom": 9.258874913727785,
+    "producer_batch_seconds": 0.09428955977352871,
+    "producer_p95_seconds": 0.2784293910008273,
+    "gpu_update_seconds": 0.9655214776178505,
+    "producer_batches_per_second": 10.605628050463595,
+    "gpu_batches_per_second": 1.0357097415038508,
+    "producer_headroom": 10.239961666348934,
     "sustainable": true,
-    "buffer_seconds": 60.115968897030044,
+    "buffer_seconds": 60.82785308992458,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400037.05241815007,
-    "upload_bytes_per_second": 5784147.848898749,
+    "generation_bytes_per_second": 418352.98070030555,
+    "upload_bytes_per_second": 5751736.850213235,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.83863906296135,
+    "estimated_full_shards_per_hour": 89.76881089932323,
     "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": 1.2689961300020514,
-      "bytes": 6434173,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3497664780006744,
       "bytes": 7459096,
@@ -562,6 +556,12 @@
       "bytes": 6081990,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9755379390007874,
+      "bytes": 2340646,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -273,3 +273,4 @@
 {"step": 584400, "loss": 0.5787742137908936, "lr": 0.0003, "grad_norm": 0.2043341006013894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85539217660, "tokens_per_second": 150159.75254509912, "data_wait_s": 9.925999620463699e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 584500, "loss": 0.5599329471588135, "lr": 0.0003, "grad_norm": 0.17117442618213818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85553788972, "tokens_per_second": 151615.05981187796, "data_wait_s": 2.068899993901141e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 584600, "loss": 0.676922619342804, "lr": 0.0003, "grad_norm": 0.1976247497956439, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85568339341, "tokens_per_second": 151083.44518003654, "data_wait_s": 1.0809999366756529e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 584700, "loss": 0.590989351272583, "lr": 0.0003, "grad_norm": 0.1806887773260374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85582950419, "tokens_per_second": 152748.95279501256, "data_wait_s": 9.019000572152436e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:886250c880ddd54566b7c1d3484b0301baf599ca726d8db7bdd39f493c38737d
-size 586530
+oid sha256:9cddcdde86685b2912cd76ad75941cc7838f0407acd279cba762f4adbbeb3ffb
+size 588648