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": 76487.22872773392,
-      "required_pages_per_second_upper_estimate": 0.005291563958206539
+      "required_token_positions_per_second": 75930.66274117952,
+      "required_pages_per_second_upper_estimate": 0.0052530594318456955
     },
     {
       "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": 9560.90359096674,
-      "required_pages_per_second_upper_estimate": 0.0040029506791673395
+      "required_token_positions_per_second": 9491.33284264744,
+      "required_pages_per_second_upper_estimate": 0.003973822859648477
     },
     {
       "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": 19121.80718193348,
-      "required_pages_per_second_upper_estimate": 0.0039825681600302665
+      "required_token_positions_per_second": 18982.66568529488,
+      "required_pages_per_second_upper_estimate": 0.003953588655688469
     },
     {
       "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": 28682.71077290022,
-      "required_pages_per_second_upper_estimate": 0.0017194847148718866
+      "required_token_positions_per_second": 28473.99852794232,
+      "required_pages_per_second_upper_estimate": 0.001706972734471806
     },
     {
       "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": 28682.71077290022,
-      "required_pages_per_second_upper_estimate": 0.002229954600681504
+      "required_token_positions_per_second": 28473.99852794232,
+      "required_pages_per_second_upper_estimate": 0.0022137281416641725
     },
     {
       "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": 28682.71077290022,
-      "required_pages_per_second_upper_estimate": 0.36791101669937815
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28473.99852794232,
+      "required_pages_per_second_upper_estimate": 0.3652338801188071
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.13483252074217233,
-    "producer_p95_seconds": 0.15249527700007093,
-    "gpu_update_seconds": 0.9639256281913972,
-    "producer_batches_per_second": 7.416608356022705,
-    "gpu_batches_per_second": 1.0374244347837174,
-    "producer_headroom": 7.149058868628751,
+    "producer_batch_seconds": 0.10179067010157539,
+    "producer_p95_seconds": 0.15287248399999953,
+    "gpu_update_seconds": 0.9709911297799729,
+    "producer_batches_per_second": 9.824083081505554,
+    "gpu_batches_per_second": 1.0298755254608767,
+    "producer_headroom": 9.539097530363396,
     "sustainable": true,
-    "buffer_seconds": 60.72731457605802,
+    "buffer_seconds": 60.201450046358325,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 452090.00362894987,
+    "generation_bytes_per_second": 449374.8202830633,
     "upload_bytes_per_second": 4116576.3146626665,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.00799066211101,
+    "estimated_full_shards_per_hour": 96.42537552231715,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -33,3 +33,4 @@
 {"step": 845800, "loss": 0.5342761278152466, "lr": 3e-05, "grad_norm": 0.2988658951245049, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123801351839, "tokens_per_second": 150000.16667407998, "data_wait_s": 1.7228000160685042e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 845900, "loss": 0.5151355862617493, "lr": 3e-05, "grad_norm": 0.2853149347511223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123816069731, "tokens_per_second": 153360.16995992142, "data_wait_s": 9.447000138607109e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 846000, "loss": 0.5326921343803406, "lr": 3e-05, "grad_norm": 0.2937797453512959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123830789717, "tokens_per_second": 152965.5423731737, "data_wait_s": 1.2217999938002322e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 846100, "loss": 0.5296584367752075, "lr": 3e-05, "grad_norm": 0.298873914583901, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123845504747, "tokens_per_second": 153350.81891504614, "data_wait_s": 9.380999927088851e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f44d372af02b114f9b8c15f0e19e6590bca533a2fba9db2c4fc5702620592e31
-size 75078
+oid sha256:1ce6cb3bf3896fc5e6f83b5cac6ee9ad13ec77fd132a58af16b7e7723d4720ea
+size 77196