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": 75840.42967725648,
-      "required_pages_per_second_upper_estimate": 0.005246816899113946
+      "required_token_positions_per_second": 76315.69423947243,
+      "required_pages_per_second_upper_estimate": 0.005279696804293761
     },
     {
       "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": 9480.05370965706,
-      "required_pages_per_second_upper_estimate": 0.003969100522200486
+      "required_token_positions_per_second": 9539.461779934054,
+      "required_pages_per_second_upper_estimate": 0.003993973440643884
     },
     {
       "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": 18960.10741931412,
-      "required_pages_per_second_upper_estimate": 0.003948890363786157
+      "required_token_positions_per_second": 19078.923559868108,
+      "required_pages_per_second_upper_estimate": 0.003973636632471214
     },
     {
       "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": 28440.16112897118,
-      "required_pages_per_second_upper_estimate": 0.0017049442340702028
+      "required_token_positions_per_second": 28618.38533980216,
+      "required_pages_per_second_upper_estimate": 0.001715628503376907
     },
     {
       "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": 28440.16112897118,
-      "required_pages_per_second_upper_estimate": 0.002211097432729155
+      "required_token_positions_per_second": 28618.38533980216,
+      "required_pages_per_second_upper_estimate": 0.002224953581195808
     },
     {
       "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": 28440.16112897118,
-      "required_pages_per_second_upper_estimate": 0.3647998502965737
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28618.38533980216,
+      "required_pages_per_second_upper_estimate": 0.3670859191108652
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08283313880475873,
-    "producer_p95_seconds": 0.17081136599972524,
-    "gpu_update_seconds": 0.9721463909652668,
-    "producer_batches_per_second": 12.07246295902227,
-    "gpu_batches_per_second": 1.0286516612041081,
-    "producer_headroom": 11.736201295675365,
+    "producer_batch_seconds": 0.5509271851874828,
+    "producer_p95_seconds": 0.16851182700065692,
+    "gpu_update_seconds": 0.9660922400659495,
+    "producer_batches_per_second": 1.8151219015625373,
+    "gpu_batches_per_second": 1.0350978493852054,
+    "producer_headroom": 1.7535751838733176,
     "sustainable": true,
-    "buffer_seconds": 60.27307623984654,
+    "buffer_seconds": 60.86381112415482,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 246227.20106864188,
-    "upload_bytes_per_second": 3603241.4732506955,
+    "generation_bytes_per_second": 228102.78095007216,
+    "upload_bytes_per_second": 4090891.094435287,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.834625473446295,
+    "estimated_full_shards_per_hour": 48.94554683090805,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 2616245,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2741041360004601,
+      "bytes": 11482818,
+      "pages": 21,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -49,3 +49,4 @@
 {"step": 724900, "loss": 0.5515720248222351, "lr": 5e-05, "grad_norm": 0.22188084885963635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106105385064, "tokens_per_second": 145296.97544987826, "data_wait_s": 1.5887999325059354e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 725000, "loss": 0.6052325963973999, "lr": 5e-05, "grad_norm": 0.21800103575122867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106120080336, "tokens_per_second": 151609.07185543582, "data_wait_s": 2.033100008702604e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 725100, "loss": 0.5414401292800903, "lr": 5e-05, "grad_norm": 0.22495387075795156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106134481815, "tokens_per_second": 110442.90534170241, "data_wait_s": 2.3059000341163483e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 725200, "loss": 0.5401269793510437, "lr": 5e-05, "grad_norm": 0.22418393062260913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106149120929, "tokens_per_second": 150576.0619204012, "data_wait_s": 1.4080000255489722e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d86622cc5af7735673fe014438197cbbb98baf7772a0a45349c00667ab81c05f
-size 108966
+oid sha256:ee5ec7eab02d015b83fa86aaeefa4429197e12ed5df0d93f50ea9dbc11f1edf7
+size 111084