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": 75910.58738063452,
-      "required_pages_per_second_upper_estimate": 0.005251670571821939
+      "required_token_positions_per_second": 75798.48320726039,
+      "required_pages_per_second_upper_estimate": 0.005243914944990396
     },
     {
       "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": 9488.823422579315,
-      "required_pages_per_second_upper_estimate": 0.003972772217868603
+      "required_token_positions_per_second": 9474.810400907549,
+      "required_pages_per_second_upper_estimate": 0.003966905258319803
     },
     {
       "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": 18977.64684515863,
-      "required_pages_per_second_upper_estimate": 0.003952543363643804
+      "required_token_positions_per_second": 18949.620801815097,
+      "required_pages_per_second_upper_estimate": 0.00394670627791181
     },
     {
       "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": 28466.470267737943,
-      "required_pages_per_second_upper_estimate": 0.0017065214267675414
+      "required_token_positions_per_second": 28424.431202722648,
+      "required_pages_per_second_upper_estimate": 0.0017040012490098099
     },
     {
       "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": 28466.470267737943,
-      "required_pages_per_second_upper_estimate": 0.0022131428525465993
+      "required_token_positions_per_second": 28424.431202722648,
+      "required_pages_per_second_upper_estimate": 0.002209874500151965
     },
     {
       "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": 28466.470267737943,
-      "required_pages_per_second_upper_estimate": 0.36513731568012137
+      "required_token_positions_per_second": 28424.431202722648,
+      "required_pages_per_second_upper_estimate": 0.3645980836921364
     }
   ],
   "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": 0.09360843096874838,
-    "producer_p95_seconds": 0.15579881799999384,
-    "gpu_update_seconds": 0.9712479186903076,
-    "producer_batches_per_second": 10.682798436540986,
-    "gpu_batches_per_second": 1.029603235956957,
-    "producer_headroom": 10.375645747278504,
+    "producer_batch_seconds": 0.13061192039063663,
+    "producer_p95_seconds": 0.15259438800012504,
+    "gpu_update_seconds": 0.9726843715117762,
+    "producer_batches_per_second": 7.656269022070734,
+    "gpu_batches_per_second": 1.02808272579292,
+    "producer_headroom": 7.447133221857953,
     "sustainable": true,
-    "buffer_seconds": 60.21737095879907,
+    "buffer_seconds": 60.30643103373013,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378881.2112837111,
+    "generation_bytes_per_second": 380382.64831930737,
     "upload_bytes_per_second": 3430023.880253648,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.29908803828715,
+    "estimated_full_shards_per_hour": 81.62126147446075,
     "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

@@ -12,3 +12,4 @@
 {"step": 843700, "loss": 0.5038239359855652, "lr": 3e-05, "grad_norm": 0.2985374037328828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123494154041, "tokens_per_second": 130223.20923862039, "data_wait_s": 1.32210000174382e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 843800, "loss": 0.5686857104301453, "lr": 3e-05, "grad_norm": 0.3051568909555626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123508825676, "tokens_per_second": 151480.6816333096, "data_wait_s": 1.8688999944060924e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 843900, "loss": 0.5337322354316711, "lr": 3e-05, "grad_norm": 0.30476746880164146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123523342733, "tokens_per_second": 149110.35123163465, "data_wait_s": 1.6943999980867375e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 844000, "loss": 0.5025510191917419, "lr": 3e-05, "grad_norm": 0.2940395860061583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123537879030, "tokens_per_second": 150556.11786133875, "data_wait_s": 1.715500002319459e-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 sha256:2f1d0a54bb7741584fb362fd18ca20af273c83fb8e72dd9ce3a7c6f29aa5efc2
-size 29817
+oid sha256:062170d8fc6c54f1c6a1fac50a44591cf8125413cd3c02227f47cd5f6524a2dc
+size 31935