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": 77202.72313327875,
-      "required_pages_per_second_upper_estimate": 0.00534106352135788
+      "required_token_positions_per_second": 77287.3755373935,
+      "required_pages_per_second_upper_estimate": 0.005346919971095177
     },
     {
       "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": 9650.340391659844,
-      "required_pages_per_second_upper_estimate": 0.004040395999964765
+      "required_token_positions_per_second": 9660.921942174187,
+      "required_pages_per_second_upper_estimate": 0.004044826274197219
     },
     {
       "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": 19300.680783319687,
-      "required_pages_per_second_upper_estimate": 0.004019822813984926
+      "required_token_positions_per_second": 19321.843884348375,
+      "required_pages_per_second_upper_estimate": 0.004024230529820696
     },
     {
       "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": 28951.02117497953,
-      "required_pages_per_second_upper_estimate": 0.0017355695137902787
+      "required_token_positions_per_second": 28982.765826522562,
+      "required_pages_per_second_upper_estimate": 0.0017374725571790072
     },
     {
       "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": 28951.02117497953,
-      "required_pages_per_second_upper_estimate": 0.0022508145542704363
+      "required_token_positions_per_second": 28982.765826522562,
+      "required_pages_per_second_upper_estimate": 0.0022532825612979414
     },
     {
       "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": 28951.02117497953,
-      "required_pages_per_second_upper_estimate": 0.37135261444798723
+      "required_token_positions_per_second": 28982.765826522562,
+      "required_pages_per_second_upper_estimate": 0.37175980075322995
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.18752708835151566,
-    "producer_p95_seconds": 0.23377709499982302,
-    "gpu_update_seconds": 0.9549922205816476,
-    "producer_batches_per_second": 5.332562931524435,
-    "gpu_batches_per_second": 1.0471289487478128,
-    "producer_headroom": 5.0925561153679,
+    "producer_batch_seconds": 0.08412893468745608,
+    "producer_p95_seconds": 0.2257737719992292,
+    "gpu_update_seconds": 0.9539462232655139,
+    "producer_batches_per_second": 11.886516853149972,
+    "gpu_batches_per_second": 1.0482771204616088,
+    "producer_headroom": 11.339097859844298,
     "sustainable": true,
-    "buffer_seconds": 60.1645098966438,
+    "buffer_seconds": 60.09861206572738,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 163,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 562992.9837057992,
+    "generation_bytes_per_second": 571464.3632394855,
     "upload_bytes_per_second": 5087388.768444568,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.4025942486786,
+    "estimated_full_shards_per_hour": 61.3114746708318,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -141,3 +141,4 @@
 {"step": 454300, "loss": 0.559476375579834, "lr": 0.0003, "grad_norm": 0.15328640335116137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66495484954, "tokens_per_second": 152184.67743569327, "data_wait_s": 1.4470999303739518e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 454400, "loss": 0.6459651589393616, "lr": 0.0003, "grad_norm": 0.17926437166238893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66510301729, "tokens_per_second": 154395.115004786, "data_wait_s": 1.0058998668682761e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 454500, "loss": 0.6405413746833801, "lr": 0.0003, "grad_norm": 0.17060514471905874, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66525073445, "tokens_per_second": 153730.21775179353, "data_wait_s": 1.4414999895961955e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 454600, "loss": 0.5783511996269226, "lr": 0.0003, "grad_norm": 0.14675936160243108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66539606627, "tokens_per_second": 151291.57431869392, "data_wait_s": 1.3990998922963627e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aecba97a8a4f35b6bd03dfe675100087d88881833512f1fdb3099a6a3bf9f6ce
-size 304605
+oid sha256:20176a3ba938bbb6f780e1134edb16f679f297da7be766623d8edc91f6089d53
+size 306723