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": 75934.00236181836,
-      "required_pages_per_second_upper_estimate": 0.005253290474550479
+      "required_token_positions_per_second": 76789.0861144831,
+      "required_pages_per_second_upper_estimate": 0.005312447152627482
     },
     {
       "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": 9491.750295227295,
-      "required_pages_per_second_upper_estimate": 0.003973997638326262
+      "required_token_positions_per_second": 9598.635764310387,
+      "required_pages_per_second_upper_estimate": 0.004018748352208946
     },
     {
       "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": 18983.50059045459,
-      "required_pages_per_second_upper_estimate": 0.0039537625444153075
+      "required_token_positions_per_second": 19197.271528620775,
+      "required_pages_per_second_upper_estimate": 0.003998285393316577
     },
     {
       "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": 28475.250885681886,
-      "required_pages_per_second_upper_estimate": 0.0017070478114060534
+      "required_token_positions_per_second": 28795.907292931166,
+      "required_pages_per_second_upper_estimate": 0.0017262706734066636
     },
     {
       "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": 28475.250885681886,
-      "required_pages_per_second_upper_estimate": 0.0022138255069698845
+      "required_token_positions_per_second": 28795.907292931166,
+      "required_pages_per_second_upper_estimate": 0.002238755132215038
     },
     {
       "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": 28475.250885681886,
-      "required_pages_per_second_upper_estimate": 0.3652499440192133
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28795.907292931166,
+      "required_pages_per_second_upper_estimate": 0.36936298011738133
     }
   ],
   "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": 0.07994828888274697,
-    "producer_p95_seconds": 0.19298305099982827,
-    "gpu_update_seconds": 0.9709484250374824,
-    "producer_batches_per_second": 12.508085088182574,
-    "gpu_batches_per_second": 1.0299208219647673,
-    "producer_headroom": 12.14470551660569,
+    "producer_batch_seconds": 0.5726594911641598,
+    "producer_p95_seconds": 0.17993605300034687,
+    "gpu_update_seconds": 0.9601364429585814,
+    "producer_batches_per_second": 1.7462384111841043,
+    "gpu_batches_per_second": 1.0415186376204848,
+    "producer_headroom": 1.6766271366719505,
     "sustainable": true,
-    "buffer_seconds": 60.19880235232391,
+    "buffer_seconds": 60.48859590639063,
     "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": 259256.10298950534,
-    "upload_bytes_per_second": 4615390.391718062,
+    "generation_bytes_per_second": 248103.39250478215,
+    "upload_bytes_per_second": 4684113.223649963,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.63032452834959,
+    "estimated_full_shards_per_hour": 53.23721248014067,
     "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": 2.719455530999767,
-      "bytes": 7662793,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.9099277899999834,
       "bytes": 1221459,
@@ -562,6 +556,12 @@
       "bytes": 6031558,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1537458799994056,
+      "bytes": 6763666,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -99,3 +99,4 @@
 {"step": 450100, "loss": 0.5955798625946045, "lr": 0.0003, "grad_norm": 0.16490552181502618, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65880555537, "tokens_per_second": 139886.8775486347, "data_wait_s": 2.092300019285176e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 450200, "loss": 0.6282391548156738, "lr": 0.0003, "grad_norm": 0.19097760285485516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65895387364, "tokens_per_second": 154523.19505666956, "data_wait_s": 1.3873001080355607e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 450300, "loss": 0.5975652933120728, "lr": 0.0003, "grad_norm": 0.16137010197765644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65909904151, "tokens_per_second": 151171.20417849618, "data_wait_s": 7.103000825736672e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 450400, "loss": 0.6156168580055237, "lr": 0.0003, "grad_norm": 0.19648303706350098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65924449840, "tokens_per_second": 121494.6019841358, "data_wait_s": 1.4375998944160528e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2cb967b2548da7988ce0225c91b03487a1e9db3c4c19975812c26119e62da59
-size 215649
+oid sha256:006c38f96f40b9973f9556ad09aa3624547038161f5af285edd04d499e40dfdd
+size 217767