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": 76916.58143182355,
-      "required_pages_per_second_upper_estimate": 0.0053212675745109884
+      "required_token_positions_per_second": 76996.0192396691,
+      "required_pages_per_second_upper_estimate": 0.005326763266378852
     },
     {
       "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": 9614.572678977944,
-      "required_pages_per_second_upper_estimate": 0.004025420805579629
+      "required_token_positions_per_second": 9624.502404958637,
+      "required_pages_per_second_upper_estimate": 0.004029578174491488
     },
     {
       "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": 19229.145357955887,
-      "required_pages_per_second_upper_estimate": 0.004004923871397676
+      "required_token_positions_per_second": 19249.004809917275,
+      "required_pages_per_second_upper_estimate": 0.004009060071512265
     },
     {
       "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": 28843.71803693383,
-      "required_pages_per_second_upper_estimate": 0.0017291368545068442
+      "required_token_positions_per_second": 28873.50721487591,
+      "required_pages_per_second_upper_estimate": 0.0017309226702390307
     },
     {
       "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": 28843.71803693383,
-      "required_pages_per_second_upper_estimate": 0.0022424722072640073
+      "required_token_positions_per_second": 28873.50721487591,
+      "required_pages_per_second_upper_estimate": 0.002244788184820258
     },
     {
       "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": 28843.71803693383,
-      "required_pages_per_second_upper_estimate": 0.3699762450062702
+      "required_token_positions_per_second": 28873.50721487591,
+      "required_pages_per_second_upper_estimate": 0.3703583485957839
     }
   ],
   "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": 0.14266716103105637,
-    "producer_p95_seconds": 0.31445397400239017,
-    "gpu_update_seconds": 0.9585449408636315,
-    "producer_batches_per_second": 7.009321505895221,
-    "gpu_batches_per_second": 1.0432479035349331,
-    "producer_headroom": 6.718749668362515,
+    "producer_batch_seconds": 0.08913320792180457,
+    "producer_p95_seconds": 0.3154512490000343,
+    "gpu_update_seconds": 0.957555997414664,
+    "producer_batches_per_second": 11.219163130281222,
+    "gpu_batches_per_second": 1.0443253477602688,
+    "producer_headroom": 10.74297694137426,
     "sustainable": true,
-    "buffer_seconds": 60.388331274408785,
+    "buffer_seconds": 60.32602783712383,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350775.7243700377,
+    "generation_bytes_per_second": 336805.063208869,
     "upload_bytes_per_second": 4952123.409171219,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.26830480886314,
+    "estimated_full_shards_per_hour": 72.27052614402345,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -358,3 +358,4 @@
 {"step": 640200, "loss": 0.5764673352241516, "lr": 0.0001, "grad_norm": 0.1832202280649628, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93708883486, "tokens_per_second": 151786.39544143138, "data_wait_s": 1.0358999134041369e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 640300, "loss": 0.5280592441558838, "lr": 0.0001, "grad_norm": 0.1931041105917027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93723538298, "tokens_per_second": 142031.75170983784, "data_wait_s": 1.463800435885787e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 640400, "loss": 0.5551453232765198, "lr": 0.0001, "grad_norm": 0.16150456430865168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93738132504, "tokens_per_second": 151911.8700938095, "data_wait_s": 1.0700001439545304e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 640500, "loss": 0.5217152237892151, "lr": 0.0001, "grad_norm": 0.1658666262688848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93752800717, "tokens_per_second": 152543.14439320512, "data_wait_s": 9.162002243101597e-06, "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 sha256:0fe0fa998558c3a0662bc6334714ebe79180f82a166b3c474ff9aaf8c185651e
-size 768909
+oid sha256:63233130667cbe1f5519245c44631db8335515fd73fa7ec47d9d558002334ef5
+size 771027