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": 77476.12572042571,
-      "required_pages_per_second_upper_estimate": 0.0053599781467181095
+      "required_token_positions_per_second": 75874.80842051648,
+      "required_pages_per_second_upper_estimate": 0.005249195300342347
     },
     {
       "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": 9684.515715053214,
-      "required_pages_per_second_upper_estimate": 0.004054704494207664
+      "required_token_positions_per_second": 9484.35105256456,
+      "required_pages_per_second_upper_estimate": 0.003970899729937131
     },
     {
       "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": 19369.031430106428,
-      "required_pages_per_second_upper_estimate": 0.004034058451182833
+      "required_token_positions_per_second": 18968.70210512912,
+      "required_pages_per_second_upper_estimate": 0.003950680410184313
     },
     {
       "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": 29053.547145159642,
-      "required_pages_per_second_upper_estimate": 0.0017417157891544844
+      "required_token_positions_per_second": 28453.05315769368,
+      "required_pages_per_second_upper_estimate": 0.0017057170915071287
     },
     {
       "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": 29053.547145159642,
-      "required_pages_per_second_upper_estimate": 0.002258785497487856
+      "required_token_positions_per_second": 28453.05315769368,
+      "required_pages_per_second_upper_estimate": 0.002212099731256288
     },
     {
       "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": 29053.547145159642,
-      "required_pages_per_second_upper_estimate": 0.37266770750964767
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28453.05315769368,
+      "required_pages_per_second_upper_estimate": 0.36496521539864396
     }
   ],
   "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": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09396749792955461,
-    "producer_p95_seconds": 0.33272527999361046,
-    "gpu_update_seconds": 0.9516221844397472,
-    "producer_batches_per_second": 10.641977513859933,
-    "gpu_batches_per_second": 1.0508372086646283,
-    "producer_headroom": 10.127141888498059,
+    "producer_batch_seconds": 0.4095926934064096,
+    "producer_p95_seconds": 0.34542785600206116,
+    "gpu_update_seconds": 0.9717059131323489,
+    "producer_batches_per_second": 2.4414498014684343,
+    "gpu_batches_per_second": 1.029117952752231,
+    "producer_headroom": 2.372371208702677,
     "sustainable": true,
-    "buffer_seconds": 60.90381980414382,
+    "buffer_seconds": 60.24576661420563,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 166,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 607021.5059575255,
+    "generation_bytes_per_second": 563524.1165151314,
     "upload_bytes_per_second": 5228690.659501593,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.12634222051774,
+    "estimated_full_shards_per_hour": 60.45957861705045,
     "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": [

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -363,3 +363,4 @@
 {"step": 681300, "loss": 0.5295174717903137, "lr": 0.0001, "grad_norm": 0.20584360366367332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99724809038, "tokens_per_second": 154015.72640290784, "data_wait_s": 1.7398000636603683e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 681400, "loss": 0.5446250438690186, "lr": 0.0001, "grad_norm": 0.19538721197666448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99739308317, "tokens_per_second": 150498.3577251656, "data_wait_s": 1.0251002095174044e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 681500, "loss": 0.5778361558914185, "lr": 0.0001, "grad_norm": 0.19999356571613733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99753818140, "tokens_per_second": 149664.32069253526, "data_wait_s": 1.180300023406744e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 681600, "loss": 0.5432363152503967, "lr": 0.0001, "grad_norm": 0.20279183690871785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99768308648, "tokens_per_second": 151271.44525978353, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08389294f97994bc5c6c462f9588383221f71b404427d6886d11c43b14196091
-size 778716
+oid sha256:ae944a660df4041069750a9d23e15a49d69c74e0b51914d12453868cfd754dd5
+size 780834