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": 77583.2545554469,
-      "required_pages_per_second_upper_estimate": 0.0053673895680980185
+      "required_token_positions_per_second": 77432.07024015207,
+      "required_pages_per_second_upper_estimate": 0.005356930286369981
     },
     {
       "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": 9697.906819430862,
-      "required_pages_per_second_upper_estimate": 0.004060311069972527
+      "required_token_positions_per_second": 9679.008780019009,
+      "required_pages_per_second_upper_estimate": 0.004052398855506722
     },
     {
       "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": 19395.813638861724,
-      "required_pages_per_second_upper_estimate": 0.0040396364789722205
+      "required_token_positions_per_second": 19358.017560038017,
+      "required_pages_per_second_upper_estimate": 0.004031764552502869
     },
     {
       "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": 29093.720458292584,
-      "required_pages_per_second_upper_estimate": 0.0017441241179357069
+      "required_token_positions_per_second": 29037.026340057026,
+      "required_pages_per_second_upper_estimate": 0.0017407253921143913
     },
     {
       "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": 29093.720458292584,
-      "required_pages_per_second_upper_estimate": 0.002261908795880213
+      "required_token_positions_per_second": 29037.026340057026,
+      "required_pages_per_second_upper_estimate": 0.00225750107755847
     },
     {
       "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": 29093.720458292584,
-      "required_pages_per_second_upper_estimate": 0.3731830076357741
+      "required_token_positions_per_second": 29037.026340057026,
+      "required_pages_per_second_upper_estimate": 0.3724557963604498
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04851420058594513,
-    "producer_p95_seconds": 0.14100640300011946,
-    "gpu_update_seconds": 0.9503081615029229,
-    "producer_batches_per_second": 20.612521445725033,
-    "gpu_batches_per_second": 1.0522902364833835,
-    "producer_headroom": 19.588247359026525,
+    "producer_batch_seconds": 0.05312440192967216,
+    "producer_p95_seconds": 0.15182420400014962,
+    "gpu_update_seconds": 0.9521636160745275,
+    "producer_batches_per_second": 18.82374132557451,
+    "gpu_batches_per_second": 1.0502396679708126,
+    "producer_headroom": 17.923281608610544,
     "sustainable": true,
-    "buffer_seconds": 60.81972233618706,
+    "buffer_seconds": 60.93847142876976,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -21,3 +21,4 @@
 {"step": 687100, "loss": 0.5114176273345947, "lr": 8e-05, "grad_norm": 0.2032273692054656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100573505138, "tokens_per_second": 154395.6483364498, "data_wait_s": 1.3889000001654495e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687200, "loss": 0.5831496119499207, "lr": 8e-05, "grad_norm": 0.2169987922337525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100588032378, "tokens_per_second": 152633.4914358772, "data_wait_s": 1.7752999610820552e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687300, "loss": 0.5587503910064697, "lr": 8e-05, "grad_norm": 0.20992206007327197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100602552012, "tokens_per_second": 152279.58006398738, "data_wait_s": 1.1000000085914508e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 687400, "loss": 0.5115691423416138, "lr": 8e-05, "grad_norm": 0.19852005288075664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100617037193, "tokens_per_second": 151616.1899802224, "data_wait_s": 1.4263000139180804e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e54d18e77bd587127fcaf38e8d93fb97311b0681a89e24edd35c70d6aac96415
-size 48879
+oid sha256:8b514532f2ac709253d36925222a04aa40b5173c0784ce9637c471cf85fac77f
+size 50997