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": 76067.98957490516,
-      "required_pages_per_second_upper_estimate": 0.00526256002084499
+      "required_token_positions_per_second": 76082.28891888399,
+      "required_pages_per_second_upper_estimate": 0.0052635492826404945
     },
     {
       "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": 9508.498696863146,
-      "required_pages_per_second_upper_estimate": 0.003981009844344794
+      "required_token_positions_per_second": 9510.286114860499,
+      "required_pages_per_second_upper_estimate": 0.003981758198934754
     },
     {
       "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": 19016.99739372629,
-      "required_pages_per_second_upper_estimate": 0.003960739045166696
+      "required_token_positions_per_second": 19020.572229720998,
+      "required_pages_per_second_upper_estimate": 0.003961483589229629
     },
     {
       "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": 28525.496090589433,
-      "required_pages_per_second_upper_estimate": 0.0017100599347202752
+      "required_token_positions_per_second": 28530.8583445815,
+      "required_pages_per_second_upper_estimate": 0.0017103813936594117
     },
     {
       "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": 28525.496090589433,
-      "required_pages_per_second_upper_estimate": 0.002217731850646146
+      "required_token_positions_per_second": 28530.8583445815,
+      "required_pages_per_second_upper_estimate": 0.0022181487423079665
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28525.496090589433,
-      "required_pages_per_second_upper_estimate": 0.3658944355586695
+      "required_token_positions_per_second": 28530.8583445815,
+      "required_pages_per_second_upper_estimate": 0.36596321679533994
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10571299378128174,
-    "producer_p95_seconds": 0.2899872490015696,
-    "gpu_update_seconds": 0.9692381829994213,
-    "producer_batches_per_second": 9.459575064811633,
-    "gpu_batches_per_second": 1.0317381398506016,
-    "producer_headroom": 9.168581347764661,
+    "producer_batch_seconds": 0.26121670339844627,
+    "producer_p95_seconds": 0.294489550000435,
+    "gpu_update_seconds": 0.9690560187878411,
+    "producer_batches_per_second": 3.8282391094824146,
+    "gpu_batches_per_second": 1.0319320871159396,
+    "producer_headroom": 3.709778150402939,
     "sustainable": true,
-    "buffer_seconds": 60.09276734596412,
+    "buffer_seconds": 60.081473164846145,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 462206.6694620903,
+    "generation_bytes_per_second": 437109.7618605049,
     "upload_bytes_per_second": 5256297.671680878,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.17879164597541,
+    "estimated_full_shards_per_hour": 93.79357950078355,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 743400, "loss": 0.5343039631843567, "lr": 5e-05, "grad_norm": 0.23291467367445792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108813587604, "tokens_per_second": 151039.99727822057, "data_wait_s": 1.371100006508641e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743500, "loss": 0.5424144268035889, "lr": 5e-05, "grad_norm": 0.24468611795184939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108828331762, "tokens_per_second": 151985.0482702344, "data_wait_s": 1.322199750575237e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743600, "loss": 0.5389540791511536, "lr": 5e-05, "grad_norm": 0.2382531738229462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108842916694, "tokens_per_second": 149502.89486380553, "data_wait_s": 1.1493000783957541e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 743700, "loss": 0.5052394866943359, "lr": 5e-05, "grad_norm": 0.2298700973910351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108857464980, "tokens_per_second": 150381.66727331592, "data_wait_s": 1.0370997188147157e-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 sha256:389cdff98fcc847967b4a3d2759e5a0f839388a37395db779eb5485a9be89d14
-size 503145
+oid sha256:77a94623105f570f8031fb55f840623542e664396efd6810343672a25924921d
+size 505263