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": 76370.36274281703,
-      "required_pages_per_second_upper_estimate": 0.005283478898203545
+      "required_token_positions_per_second": 77418.02290610711,
+      "required_pages_per_second_upper_estimate": 0.0053559584592064455
     },
     {
       "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": 9546.295342852129,
-      "required_pages_per_second_upper_estimate": 0.003996834510736661
+      "required_token_positions_per_second": 9677.25286326339,
+      "required_pages_per_second_upper_estimate": 0.004051663689828856
     },
     {
       "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": 19092.590685704257,
-      "required_pages_per_second_upper_estimate": 0.003976483134356538
+      "required_token_positions_per_second": 19354.50572652678,
+      "required_pages_per_second_upper_estimate": 0.004031033130195756
     },
     {
       "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": 28638.88602855639,
-      "required_pages_per_second_upper_estimate": 0.0017168574883649839
+      "required_token_positions_per_second": 29031.758589790166,
+      "required_pages_per_second_upper_estimate": 0.0017404095985292822
     },
     {
       "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": 28638.88602855639,
-      "required_pages_per_second_upper_estimate": 0.0022265474195734495
+      "required_token_positions_per_second": 29031.758589790166,
+      "required_pages_per_second_upper_estimate": 0.0022570915331456064
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28638.88602855639,
-      "required_pages_per_second_upper_estimate": 0.36734887993427984
+      "required_token_positions_per_second": 29031.758589790166,
+      "required_pages_per_second_upper_estimate": 0.37238822731609605
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11466585946089936,
-    "producer_p95_seconds": 0.2418848990000697,
-    "gpu_update_seconds": 0.9654006783794469,
-    "producer_batches_per_second": 8.720991624721536,
-    "gpu_batches_per_second": 1.0358393384171147,
-    "producer_headroom": 8.419251230647646,
+    "producer_batch_seconds": 0.09553327954678537,
+    "producer_p95_seconds": 0.22782008300055168,
+    "gpu_update_seconds": 0.9523363841184321,
+    "producer_batches_per_second": 10.467556486535894,
+    "gpu_batches_per_second": 1.0500491388089614,
+    "producer_headroom": 9.968634894943033,
     "sustainable": true,
-    "buffer_seconds": 60.820242737905154,
+    "buffer_seconds": 60.94952858357966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 474838.37521415454,
-    "upload_bytes_per_second": 1872218.810960124,
+    "generation_bytes_per_second": 475684.75345751684,
+    "upload_bytes_per_second": 1859691.174697499,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.88926164930798,
+    "estimated_full_shards_per_hour": 102.07087471765641,
     "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": [
-    {
-      "seconds": 1.190589429001193,
-      "bytes": 6346216,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.0324392470010935,
       "bytes": 4954714,
@@ -562,6 +556,12 @@
       "bytes": 11160724,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 0.9322519389988884,
+      "bytes": 2962222,
+      "pages": 46,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -160,3 +160,4 @@
 {"step": 661000, "loss": 0.5415875315666199, "lr": 0.0001, "grad_norm": 0.19328856622939006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96752770905, "tokens_per_second": 152089.57396724008, "data_wait_s": 8.262999472208321e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 661100, "loss": 0.5500502586364746, "lr": 0.0001, "grad_norm": 0.19395241358893767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96767415798, "tokens_per_second": 152284.56853286334, "data_wait_s": 1.1494001228129491e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 661200, "loss": 0.5655838847160339, "lr": 0.0001, "grad_norm": 0.18529066629640334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96782162489, "tokens_per_second": 154032.61333851627, "data_wait_s": 8.934999641496688e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 661300, "loss": 0.501309871673584, "lr": 0.0001, "grad_norm": 0.1910924292280799, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96796881185, "tokens_per_second": 153321.34276279656, "data_wait_s": 1.4727000234415755e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f876269aa8bee2dd6998550391c667204ec249560ea14b5d111073dac237b20
-size 345630
+oid sha256:b479f18cea19a69778c728bd7da1f0fb93940d378ad61176e96d457395710537
+size 347748