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": 77344.8278782979,
-      "required_pages_per_second_upper_estimate": 0.0053508946573467405
+      "required_token_positions_per_second": 77572.38603714197,
+      "required_pages_per_second_upper_estimate": 0.005366637658783247
     },
     {
       "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": 9668.103484787238,
-      "required_pages_per_second_upper_estimate": 0.00404783303612164
+      "required_token_positions_per_second": 9696.548254642747,
+      "required_pages_per_second_upper_estimate": 0.004059742267265802
     },
     {
       "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": 19336.206969574476,
-      "required_pages_per_second_upper_estimate": 0.004027221981693264
+      "required_token_positions_per_second": 19393.096509285493,
+      "required_pages_per_second_upper_estimate": 0.0040390705725370174
     },
     {
       "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": 29004.310454361712,
-      "required_pages_per_second_upper_estimate": 0.0017387641247212223
+      "required_token_positions_per_second": 29089.64476392824,
+      "required_pages_per_second_upper_estimate": 0.0017438797862817895
     },
     {
       "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": 29004.310454361712,
-      "required_pages_per_second_upper_estimate": 0.002254957561347632
+      "required_token_positions_per_second": 29089.64476392824,
+      "required_pages_per_second_upper_estimate": 0.00226159192857047
     },
     {
       "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": 29004.310454361712,
-      "required_pages_per_second_upper_estimate": 0.3720361521063315
+      "required_token_positions_per_second": 29089.64476392824,
+      "required_pages_per_second_upper_estimate": 0.373130729004608
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 0.0473466057421863,
-    "producer_p95_seconds": 0.1212602569999035,
-    "gpu_update_seconds": 0.9532376245766688,
-    "producer_batches_per_second": 21.12083821689862,
-    "gpu_batches_per_second": 1.0490563677069484,
-    "producer_headroom": 20.133177650944567,
+    "producer_batch_seconds": 0.045167116648418215,
+    "producer_p95_seconds": 0.1250824300000204,
+    "gpu_update_seconds": 0.9504413073577334,
+    "producer_batches_per_second": 22.14000082812505,
+    "gpu_batches_per_second": 1.0521428227693952,
+    "producer_headroom": 21.042771331984472,
     "sustainable": true,
-    "buffer_seconds": 60.05397034833014,
+    "buffer_seconds": 60.828243670894935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 64,
+    "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -9,3 +9,4 @@
 {"step": 645900, "loss": 0.5465204119682312, "lr": 0.0001, "grad_norm": 0.19122648496474157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94543151532, "tokens_per_second": 154011.2369638154, "data_wait_s": 1.0971999927278375e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 646000, "loss": 0.6286115050315857, "lr": 0.0001, "grad_norm": 0.19377578634339912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94557920764, "tokens_per_second": 154680.50080561815, "data_wait_s": 1.4688999954159954e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 646100, "loss": 0.5347440242767334, "lr": 0.0001, "grad_norm": 0.17545354614213696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94572621539, "tokens_per_second": 154321.7722860928, "data_wait_s": 1.0644999974829261e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 646200, "loss": 0.5362868309020996, "lr": 0.0001, "grad_norm": 0.19484272820118434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94587097212, "tokens_per_second": 151959.3742421955, "data_wait_s": 1.7045000049620285e-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:c9c39cfdcbeea9195a3869d4647c7786bd2eb4e9394f744c884a9b5a2fa67fde
-size 23463
+oid sha256:43f534bae55dcb97a54e954c6e10ed2f30984464f41693af16bff290e97ba925
+size 25581