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": 76300.87610840867,
-      "required_pages_per_second_upper_estimate": 0.005278671651603969
+      "required_token_positions_per_second": 76787.68560700245,
+      "required_pages_per_second_upper_estimate": 0.005312350262270346
     },
     {
       "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": 9537.609513551084,
-      "required_pages_per_second_upper_estimate": 0.003993197935389055
+      "required_token_positions_per_second": 9598.460700875306,
+      "required_pages_per_second_upper_estimate": 0.004018675056804417
     },
     {
       "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": 19075.21902710217,
-      "required_pages_per_second_upper_estimate": 0.003972865075991167
+      "required_token_positions_per_second": 19196.921401750613,
+      "required_pages_per_second_upper_estimate": 0.003998212471122989
     },
     {
       "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": 28612.828540653252,
-      "required_pages_per_second_upper_estimate": 0.001715295381752671
+      "required_token_positions_per_second": 28795.382102625917,
+      "required_pages_per_second_upper_estimate": 0.001726239189049783
     },
     {
       "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": 28612.828540653252,
-      "required_pages_per_second_upper_estimate": 0.002224521564503757
+      "required_token_positions_per_second": 28795.382102625917,
+      "required_pages_per_second_upper_estimate": 0.002238714300978873
     },
     {
       "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": 28612.828540653252,
-      "required_pages_per_second_upper_estimate": 0.36701464245780907
+      "required_token_positions_per_second": 28795.382102625917,
+      "required_pages_per_second_upper_estimate": 0.36935624354005103
     }
   ],
   "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": 0.08601940082030524,
-    "producer_p95_seconds": 0.2970600000007835,
-    "gpu_update_seconds": 0.9662798615214704,
-    "producer_batches_per_second": 11.625284417976856,
-    "gpu_batches_per_second": 1.0348968656196922,
-    "producer_headroom": 11.233278217550383,
+    "producer_batch_seconds": 0.12995581200766537,
+    "producer_p95_seconds": 0.3062060800002655,
+    "gpu_update_seconds": 0.9601539545981129,
+    "producer_batches_per_second": 7.694923255460214,
+    "gpu_batches_per_second": 1.0414996420220601,
+    "producer_headroom": 7.388310994059109,
     "sustainable": true,
-    "buffer_seconds": 60.87563127585263,
+    "buffer_seconds": 60.489699139681115,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 532379.1971722487,
-    "upload_bytes_per_second": 5252930.623792579,
+    "generation_bytes_per_second": 532605.5665549352,
+    "upload_bytes_per_second": 5224895.868396162,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.23618255973432,
+    "estimated_full_shards_per_hour": 114.28475615964929,
     "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": 1.2808730480028316,
-      "bytes": 10383198,
-      "pages": 27,
-      "shards": 3
-    },
     {
       "seconds": 1.2442956449995108,
       "bytes": 13618701,
@@ -562,6 +556,12 @@
       "bytes": 1231448,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.1650056679973204,
+      "bytes": 7481151,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -276,3 +276,4 @@
 {"step": 632000, "loss": 0.5255146622657776, "lr": 0.0001, "grad_norm": 0.16012158731824785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92507969610, "tokens_per_second": 151008.07078913908, "data_wait_s": 1.7130001651821658e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 632100, "loss": 0.5381714105606079, "lr": 0.0001, "grad_norm": 0.18200650515160238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92522667483, "tokens_per_second": 152262.84686322522, "data_wait_s": 1.3706998288398609e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 632200, "loss": 0.5619529485702515, "lr": 0.0001, "grad_norm": 0.16172479192987982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92537393479, "tokens_per_second": 151958.09665879665, "data_wait_s": 1.5820001863176003e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 632300, "loss": 0.5189945101737976, "lr": 0.0001, "grad_norm": 0.16577218858353351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92552137809, "tokens_per_second": 152821.69150754684, "data_wait_s": 1.4707002264913172e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39d0aebf51d24ad3a2ce5e84c203f9b06ac18dad3675d5a17183597824bdc64c
-size 593667
+oid sha256:05581be2e0f9c6011e4aee6acf400cb12b6053e01afc1ee6acf51708b1522a1f
+size 595785