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": 75348.33365957985,
-      "required_pages_per_second_upper_estimate": 0.005212772554791003
+      "required_token_positions_per_second": 76932.71991042611,
+      "required_pages_per_second_upper_estimate": 0.005322384071907143
     },
     {
       "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": 9418.54170744748,
-      "required_pages_per_second_upper_estimate": 0.003943346731392008
+      "required_token_positions_per_second": 9616.589988803264,
+      "required_pages_per_second_upper_estimate": 0.004026265411077271
     },
     {
       "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": 18837.08341489496,
-      "required_pages_per_second_upper_estimate": 0.003923267708026816
+      "required_token_positions_per_second": 19233.179977606527,
+      "required_pages_per_second_upper_estimate": 0.004005764176270829
     },
     {
       "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": 28255.625122342444,
-      "required_pages_per_second_upper_estimate": 0.001693881582242977
+      "required_token_positions_per_second": 28849.76996640979,
+      "required_pages_per_second_upper_estimate": 0.0017294996584381666
     },
     {
       "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": 28255.625122342444,
-      "required_pages_per_second_upper_estimate": 0.002196750569901882
+      "required_token_positions_per_second": 28849.76996640979,
+      "required_pages_per_second_upper_estimate": 0.0022429427181611386
     },
     {
       "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": 28255.625122342444,
-      "required_pages_per_second_upper_estimate": 0.36243282054286685
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28849.76996640979,
+      "required_pages_per_second_upper_estimate": 0.3700538726595322
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.3525562218593308,
-    "producer_p95_seconds": 0.20866720499907387,
-    "gpu_update_seconds": 0.9784954280886895,
-    "producer_batches_per_second": 2.836427037725058,
-    "gpu_batches_per_second": 1.0219771817976868,
-    "producer_headroom": 2.775430888521114,
+    "producer_batch_seconds": 0.05517328707033187,
+    "producer_p95_seconds": 0.19022394199964765,
+    "gpu_update_seconds": 0.9583438631292718,
+    "producer_batches_per_second": 18.124713119326298,
+    "gpu_batches_per_second": 1.0434667956600763,
+    "producer_headroom": 17.369707588884957,
     "sustainable": true,
-    "buffer_seconds": 60.66671654149875,
+    "buffer_seconds": 60.37566337714412,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329775.5111941738,
+    "generation_bytes_per_second": 318282.3104578806,
     "upload_bytes_per_second": 5062401.148541981,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.76214792126571,
+    "estimated_full_shards_per_hour": 68.29597459127726,
     "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

@@ -99,3 +99,4 @@
 {"step": 407000, "loss": 0.572537362575531, "lr": 0.0003, "grad_norm": 0.13790204543994558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59574217989, "tokens_per_second": 152536.78562929915, "data_wait_s": 1.2856000466854312e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 407100, "loss": 0.5529351830482483, "lr": 0.0003, "grad_norm": 0.16667881099229423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59588873855, "tokens_per_second": 146776.5069391976, "data_wait_s": 8.43499947222881e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 407200, "loss": 0.626397430896759, "lr": 0.0003, "grad_norm": 0.17565449531304742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59603431026, "tokens_per_second": 151187.81257461925, "data_wait_s": 8.783999874140136e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 407300, "loss": 0.5221507549285889, "lr": 0.0003, "grad_norm": 0.1611347300725007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59618026986, "tokens_per_second": 151467.20022939064, "data_wait_s": 1.884400080598425e-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 sha256:63559fc95aec5937ba0621748e5b282c5d553dc251e83fd340a19e587a090170
-size 215649
+oid sha256:3712edd9f75a82d34fdbc0ca789c672805e4112fdeb3d57e1d2274c0b17322d0
+size 217767