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": 76832.94365758017,
-      "required_pages_per_second_upper_estimate": 0.005315481319222458
+      "required_token_positions_per_second": 75847.33368450179,
+      "required_pages_per_second_upper_estimate": 0.005247294534354681
     },
     {
       "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": 9604.11795719752,
-      "required_pages_per_second_upper_estimate": 0.0040210436318895825
+      "required_token_positions_per_second": 9480.916710562724,
+      "required_pages_per_second_upper_estimate": 0.003969461842658179
     },
     {
       "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": 19208.23591439504,
-      "required_pages_per_second_upper_estimate": 0.00400056898572306
+      "required_token_positions_per_second": 18961.833421125448,
+      "required_pages_per_second_upper_estimate": 0.003949249844445727
     },
     {
       "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": 28812.35387159256,
-      "required_pages_per_second_upper_estimate": 0.0017272566206849426
+      "required_token_positions_per_second": 28442.75013168817,
+      "required_pages_per_second_upper_estimate": 0.0017050994408299082
     },
     {
       "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": 28812.35387159256,
-      "required_pages_per_second_upper_estimate": 0.002240033781365107
+      "required_token_positions_per_second": 28442.75013168817,
+      "required_pages_per_second_upper_estimate": 0.002211298716302582
     },
     {
       "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": 28812.35387159256,
-      "required_pages_per_second_upper_estimate": 0.3695739391694894
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28442.75013168817,
+      "required_pages_per_second_upper_estimate": 0.36483305924357273
     }
   ],
   "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": 0.2938680931798672,
-    "producer_p95_seconds": 0.28481914599979063,
-    "gpu_update_seconds": 0.959588380845879,
-    "producer_batches_per_second": 3.4028872926600178,
-    "gpu_batches_per_second": 1.0421134936195227,
-    "producer_headroom": 3.2653711073646434,
+    "producer_batch_seconds": 0.10584492671901558,
+    "producer_p95_seconds": 0.2781709619994217,
+    "gpu_update_seconds": 0.9720579012926482,
+    "producer_batches_per_second": 9.447783951465905,
+    "gpu_batches_per_second": 1.0287453027954345,
+    "producer_headroom": 9.18379303972831,
     "sustainable": true,
-    "buffer_seconds": 60.45406799329038,
+    "buffer_seconds": 60.26758988014419,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339885.2972762095,
+    "generation_bytes_per_second": 339654.93323070806,
     "upload_bytes_per_second": 4832319.261631986,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.93147267069544,
+    "estimated_full_shards_per_hour": 72.88204190913135,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -288,3 +288,4 @@
 {"step": 788800, "loss": 0.5539882779121399, "lr": 3e-05, "grad_norm": 0.26287225374246087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115457688429, "tokens_per_second": 143153.21057854767, "data_wait_s": 1.064100069925189e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788900, "loss": 0.5504850745201111, "lr": 3e-05, "grad_norm": 0.26454972272552435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115472317052, "tokens_per_second": 151626.21056267782, "data_wait_s": 1.154200072051026e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 789000, "loss": 0.5050582885742188, "lr": 3e-05, "grad_norm": 0.2650915935230674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115486997166, "tokens_per_second": 151954.43834289283, "data_wait_s": 9.687999408924952e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 789100, "loss": 0.45617660880088806, "lr": 3e-05, "grad_norm": 0.2658275310719928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115501707800, "tokens_per_second": 152874.7725302047, "data_wait_s": 8.409002475673333e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f180ff8800a4e5f315de3ec831fa8c5bb3039569819ccdba962a1bcf5aad26a8
-size 618300
+oid sha256:4110b1e0f44c63020d5103535ed286466d0af5edb0f29d5a88f2bb0d34b9f0d3
+size 620418