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": 76523.0337619229,
-      "required_pages_per_second_upper_estimate": 0.0052940410335508585
+      "required_token_positions_per_second": 76188.9364639213,
+      "required_pages_per_second_upper_estimate": 0.005270927407262573
     },
     {
       "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": 9565.379220240362,
-      "required_pages_per_second_upper_estimate": 0.004004824531682437
+      "required_token_positions_per_second": 9523.617057990163,
+      "required_pages_per_second_upper_estimate": 0.003987339586441396
     },
     {
       "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": 19130.758440480724,
-      "required_pages_per_second_upper_estimate": 0.003984432471125095
+      "required_token_positions_per_second": 19047.234115980325,
+      "required_pages_per_second_upper_estimate": 0.003967036557016223
     },
     {
       "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": 28696.137660721088,
-      "required_pages_per_second_upper_estimate": 0.001720289636295077
+      "required_token_positions_per_second": 28570.851173970488,
+      "required_pages_per_second_upper_estimate": 0.001712778902715767
     },
     {
       "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": 28696.137660721088,
-      "required_pages_per_second_upper_estimate": 0.002230998482150938
+      "required_token_positions_per_second": 28570.851173970488,
+      "required_pages_per_second_upper_estimate": 0.002221258008883095
     },
     {
       "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": 28696.137660721088,
-      "required_pages_per_second_upper_estimate": 0.3680832423996753
+      "required_token_positions_per_second": 28570.851173970488,
+      "required_pages_per_second_upper_estimate": 0.3664762018697873
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11609044850001737,
-    "producer_p95_seconds": 0.3214824280003086,
-    "gpu_update_seconds": 0.9634746085653274,
-    "producer_batches_per_second": 8.61397309529604,
-    "gpu_batches_per_second": 1.0379100716406644,
-    "producer_headroom": 8.299344356182612,
+    "producer_batch_seconds": 0.14993500997690035,
+    "producer_p95_seconds": 0.3327517670040834,
+    "gpu_update_seconds": 0.967699556154237,
+    "producer_batches_per_second": 6.669556364147803,
+    "gpu_batches_per_second": 1.033378587021502,
+    "producer_headroom": 6.454126733331496,
     "sustainable": true,
-    "buffer_seconds": 60.698900339615626,
+    "buffer_seconds": 60.96507203771693,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 234818.15102500288,
+    "generation_bytes_per_second": 226320.4537732936,
     "upload_bytes_per_second": 3763872.806813212,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.38650892317357,
+    "estimated_full_shards_per_hour": 48.563100909224566,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -380,3 +380,4 @@
 {"step": 880500, "loss": 0.561471700668335, "lr": 3e-05, "grad_norm": 0.3201716198123726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128880287924, "tokens_per_second": 118832.45793314082, "data_wait_s": 1.4842997188679874e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 880600, "loss": 0.5296868085861206, "lr": 3e-05, "grad_norm": 0.31415713754224395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128895095996, "tokens_per_second": 153974.0727408594, "data_wait_s": 1.1980999261140823e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 880700, "loss": 0.6092408299446106, "lr": 3e-05, "grad_norm": 0.3469923657716512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128909907879, "tokens_per_second": 153989.6833708124, "data_wait_s": 1.396799780195579e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 880800, "loss": 0.5035049915313721, "lr": 3e-05, "grad_norm": 0.29853296905549626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128924604752, "tokens_per_second": 153178.8784803436, "data_wait_s": 1.2065000191796571e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39b8ab10af1f70b9488e48b29f674090fba784f99eeea2e466cb4f8814c7ef2c
-size 815505
+oid sha256:551ee4f4535bdf0e5f0fcdd99adaa2ec4890011cc4de2a06a0b4716cc1e9d8e6
+size 817623