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": 76367.54238620421,
-      "required_pages_per_second_upper_estimate": 0.005283283779388941
+      "required_token_positions_per_second": 76433.6275587245,
+      "required_pages_per_second_upper_estimate": 0.005287855704962621
     },
     {
       "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": 9545.942798275526,
-      "required_pages_per_second_upper_estimate": 0.003996686907684405
+      "required_token_positions_per_second": 9554.203444840563,
+      "required_pages_per_second_upper_estimate": 0.004000145467899271
     },
     {
       "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": 19091.885596551052,
-      "required_pages_per_second_upper_estimate": 0.003976336282880377
+      "required_token_positions_per_second": 19108.406889681126,
+      "required_pages_per_second_upper_estimate": 0.003979777232543572
     },
     {
       "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": 28637.828394826574,
-      "required_pages_per_second_upper_estimate": 0.0017167940848377948
+      "required_token_positions_per_second": 28662.61033452169,
+      "required_pages_per_second_upper_estimate": 0.001718279724282686
     },
     {
       "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": 28637.828394826574,
-      "required_pages_per_second_upper_estimate": 0.0022264651932029957
+      "required_token_positions_per_second": 28662.61033452169,
+      "required_pages_per_second_upper_estimate": 0.002228391880010058
     },
     {
       "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": 28637.828394826574,
-      "required_pages_per_second_upper_estimate": 0.3673353137443924
+      "required_token_positions_per_second": 28662.61033452169,
+      "required_pages_per_second_upper_estimate": 0.3676531898580276
     }
   ],
   "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": 0.07650065592190458,
-    "producer_p95_seconds": 0.28155908600092516,
-    "gpu_update_seconds": 0.9654363319320194,
-    "producer_batches_per_second": 13.071783345502899,
-    "gpu_batches_per_second": 1.035801084882327,
-    "producer_headroom": 12.61997456489238,
+    "producer_batch_seconds": 0.40982678061729416,
+    "producer_p95_seconds": 0.29209410100156674,
+    "gpu_update_seconds": 0.9646016073665252,
+    "producer_batches_per_second": 2.440055280169266,
+    "gpu_batches_per_second": 1.0366974224002348,
+    "producer_headroom": 2.353681245314451,
     "sustainable": true,
-    "buffer_seconds": 60.82248891171722,
+    "buffer_seconds": 60.76990126409109,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323104.31627050944,
+    "generation_bytes_per_second": 299804.07934916776,
     "upload_bytes_per_second": 4909392.785882454,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.33066478811705,
+    "estimated_full_shards_per_hour": 64.3309763465526,
     "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

@@ -254,3 +254,4 @@
 {"step": 867900, "loss": 0.4677457809448242, "lr": 3e-05, "grad_norm": 0.3257396143011411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127036817773, "tokens_per_second": 150672.52302423256, "data_wait_s": 9.336003131465986e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 868000, "loss": 0.5193307399749756, "lr": 3e-05, "grad_norm": 0.30677056004051867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127051330628, "tokens_per_second": 150989.08953530111, "data_wait_s": 1.0551000741543248e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 868100, "loss": 0.5418099761009216, "lr": 3e-05, "grad_norm": 0.3158439129922565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127065902252, "tokens_per_second": 150349.03504273325, "data_wait_s": 1.3627002772409469e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 868200, "loss": 0.5475059747695923, "lr": 3e-05, "grad_norm": 0.3061291304394692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127080629656, "tokens_per_second": 152992.04779714282, "data_wait_s": 1.075999898603186e-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:bb80fd42116c78be5013922b48f7749de4f2d91ebcc280fb34f9323b1a3ed033
-size 546288
+oid sha256:dd6a892b4481e9015d8f364a8d6b74f30d57fc9fac9058c0f9223530dd9c4027
+size 548406