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": 77229.96364509674,
-      "required_pages_per_second_upper_estimate": 0.005342948083172144
+      "required_token_positions_per_second": 75948.92028870722,
+      "required_pages_per_second_upper_estimate": 0.0052543225313470086
     },
     {
       "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": 9653.745455637092,
-      "required_pages_per_second_upper_estimate": 0.004041821629146218
+      "required_token_positions_per_second": 9493.615036088402,
+      "required_pages_per_second_upper_estimate": 0.003974778366384589
     },
     {
       "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": 19307.490911274184,
-      "required_pages_per_second_upper_estimate": 0.004021241184042696
+      "required_token_positions_per_second": 18987.230072176804,
+      "required_pages_per_second_upper_estimate": 0.003954539297104998
     },
     {
       "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": 28961.236366911275,
-      "required_pages_per_second_upper_estimate": 0.001736181898948892
+      "required_token_positions_per_second": 28480.84510826521,
+      "required_pages_per_second_upper_estimate": 0.0017073831765080404
     },
     {
       "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": 28961.236366911275,
-      "required_pages_per_second_upper_estimate": 0.0022516087405112482
+      "required_token_positions_per_second": 28480.84510826521,
+      "required_pages_per_second_upper_estimate": 0.0022142604331693533
     },
     {
       "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": 28961.236366911275,
-      "required_pages_per_second_upper_estimate": 0.37148364396186906
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28480.84510826521,
+      "required_pages_per_second_upper_estimate": 0.36532170069990394
     }
   ],
   "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": 0.07064870839070636,
-    "producer_p95_seconds": 0.18770493999909377,
-    "gpu_update_seconds": 0.9546553762320841,
-    "producer_batches_per_second": 14.154540440707438,
-    "gpu_batches_per_second": 1.047498421835622,
-    "producer_headroom": 13.512708129815808,
+    "producer_batch_seconds": 0.291002199968716,
+    "producer_p95_seconds": 0.17755140400004166,
+    "gpu_update_seconds": 0.9707577108369051,
+    "producer_batches_per_second": 3.436400137550521,
+    "gpu_batches_per_second": 1.0301231592977869,
+    "producer_headroom": 3.33591193104817,
     "sustainable": true,
-    "buffer_seconds": 60.1432887026213,
+    "buffer_seconds": 60.18697807188811,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 442617.6009660003,
+    "generation_bytes_per_second": 405961.4784004424,
     "upload_bytes_per_second": 5130396.743258689,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.97543355689055,
+    "estimated_full_shards_per_hour": 87.10988296518282,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -83,3 +83,4 @@
 {"step": 405400, "loss": 0.58574378490448, "lr": 0.0003, "grad_norm": 0.15193545818179702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59340087934, "tokens_per_second": 150505.42132527204, "data_wait_s": 1.2100999811082147e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 405500, "loss": 0.5842244029045105, "lr": 0.0003, "grad_norm": 0.14365073111179408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59354577552, "tokens_per_second": 150671.975644508, "data_wait_s": 1.2495000191847794e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 405600, "loss": 0.5392022728919983, "lr": 0.0003, "grad_norm": 0.1568557557438453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59369203489, "tokens_per_second": 151852.07231751402, "data_wait_s": 9.295999916503206e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 405700, "loss": 0.6061990261077881, "lr": 0.0003, "grad_norm": 0.1505809875030439, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59383779968, "tokens_per_second": 151296.8658098422, "data_wait_s": 1.0068999472423457e-05, "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 sha256:88b55a60f474aba6b1538a98c5e3ca93ae32bef0976ab6a4cde3a4f6e4495d91
-size 181761
+oid sha256:edad9a0511a32380dfd4ed891429d82608cbb819b8206432e06373a53dc138f3
+size 183879