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": 77048.20933599083,
-      "required_pages_per_second_upper_estimate": 0.005330373898340097
+      "required_token_positions_per_second": 77365.33532253442,
+      "required_pages_per_second_upper_estimate": 0.005352313409922849
     },
     {
       "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": 9631.026166998854,
-      "required_pages_per_second_upper_estimate": 0.004032309537426084
+      "required_token_positions_per_second": 9670.666915316802,
+      "required_pages_per_second_upper_estimate": 0.004048906290953853
     },
     {
       "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": 19262.052333997708,
-      "required_pages_per_second_upper_estimate": 0.004011777526691871
+      "required_token_positions_per_second": 19341.333830633605,
+      "required_pages_per_second_upper_estimate": 0.004028289771647462
     },
     {
       "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": 28893.078500996562,
-      "required_pages_per_second_upper_estimate": 0.0017320959389583402
+      "required_token_positions_per_second": 29012.00074595041,
+      "required_pages_per_second_upper_estimate": 0.0017392251459595694
     },
     {
       "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": 28893.078500996562,
-      "required_pages_per_second_upper_estimate": 0.0022463097662311486
+      "required_token_positions_per_second": 29012.00074595041,
+      "required_pages_per_second_upper_estimate": 0.002255555447692636
     },
     {
       "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": 28893.078500996562,
-      "required_pages_per_second_upper_estimate": 0.37060938804012983
+      "required_token_positions_per_second": 29012.00074595041,
+      "required_pages_per_second_upper_estimate": 0.3721347949096395
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06663329815634711,
-    "producer_p95_seconds": 0.239838809999128,
-    "gpu_update_seconds": 0.956907378320603,
-    "producer_batches_per_second": 15.007511674622783,
-    "gpu_batches_per_second": 1.0450332212455353,
-    "producer_headroom": 14.36079865167913,
+    "producer_batch_seconds": 0.10793024863289702,
+    "producer_p95_seconds": 0.2442270199990162,
+    "gpu_update_seconds": 0.9529849472328861,
+    "producer_batches_per_second": 9.265243179428767,
+    "gpu_batches_per_second": 1.0493345177210072,
+    "producer_headroom": 8.82963728244778,
     "sustainable": true,
-    "buffer_seconds": 60.28516483419799,
+    "buffer_seconds": 60.038051675671824,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362530.7025558989,
-    "upload_bytes_per_second": 5297331.82625389,
+    "generation_bytes_per_second": 398086.69224948075,
+    "upload_bytes_per_second": 5232468.799184691,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.79064948566176,
+    "estimated_full_shards_per_hour": 85.42013717282597,
     "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": 1.23988030000055,
-      "bytes": 8293432,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.1932894399997167,
       "bytes": 6531618,
@@ -562,6 +556,12 @@
       "bytes": 6119662,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8502458040002239,
+      "bytes": 891680,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -147,3 +147,4 @@
 {"step": 899700, "loss": 0.5507715344429016, "lr": 3e-05, "grad_norm": 0.3216003760680233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131690786780, "tokens_per_second": 153708.08054490155, "data_wait_s": 1.6756999684730545e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 899800, "loss": 0.588070809841156, "lr": 3e-05, "grad_norm": 0.340170300486688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131705363330, "tokens_per_second": 151823.2719513825, "data_wait_s": 1.5900999642326497e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 899900, "loss": 0.5504563450813293, "lr": 3e-05, "grad_norm": 0.32135612538305525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131719837800, "tokens_per_second": 151262.79724883498, "data_wait_s": 1.4798999472986907e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 900000, "loss": 0.5382843017578125, "lr": 3e-05, "grad_norm": 0.327387548139761, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131734420906, "tokens_per_second": 134785.8818095809, "data_wait_s": 8.85099871084094e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93aa1715f95dad7dc49abdb7a2b2ccddaded62244375d24648a55fa34e2f9ed7
-size 317313
+oid sha256:872cd2bfd970266601912acc975e461b17ecbd85697f0b57ab664174d4655fa1
+size 319431