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": 77295.09814462098,
-      "required_pages_per_second_upper_estimate": 0.005347454238982092
+      "required_token_positions_per_second": 77380.64120054741,
+      "required_pages_per_second_upper_estimate": 0.005353372306078318
     },
     {
       "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": 9661.887268077622,
-      "required_pages_per_second_upper_estimate": 0.00404523043599469
+      "required_token_positions_per_second": 9672.580150068427,
+      "required_pages_per_second_upper_estimate": 0.0040497073223914725
     },
     {
       "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": 19323.774536155244,
-      "required_pages_per_second_upper_estimate": 0.004024632633677355
+      "required_token_positions_per_second": 19345.160300136853,
+      "required_pages_per_second_upper_estimate": 0.004029086724334202
     },
     {
       "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": 28985.661804232866,
-      "required_pages_per_second_upper_estimate": 0.001737646166620838
+      "required_token_positions_per_second": 29017.740450205278,
+      "required_pages_per_second_upper_estimate": 0.0017395692324656277
     },
     {
       "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": 28985.661804232866,
-      "required_pages_per_second_upper_estimate": 0.0022535077108268575
+      "required_token_positions_per_second": 29017.740450205278,
+      "required_pages_per_second_upper_estimate": 0.002256001684452161
     },
     {
       "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": 28985.661804232866,
-      "required_pages_per_second_upper_estimate": 0.37179694724583917
+      "required_token_positions_per_second": 29017.740450205278,
+      "required_pages_per_second_upper_estimate": 0.3722084176730067
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05129268785157315,
-    "producer_p95_seconds": 0.12482571999998981,
-    "gpu_update_seconds": 0.9538509138322478,
-    "producer_batches_per_second": 19.495956283159178,
-    "gpu_batches_per_second": 1.0483818650257837,
-    "producer_headroom": 18.596235716724937,
+    "producer_batch_seconds": 0.04598225053126015,
+    "producer_p95_seconds": 0.1300537820000045,
+    "gpu_update_seconds": 0.9527964469681653,
+    "producer_batches_per_second": 21.747521890433987,
+    "gpu_batches_per_second": 1.0495421169779109,
+    "producer_headroom": 20.720961587567903,
     "sustainable": true,
-    "buffer_seconds": 60.092607571431614,
+    "buffer_seconds": 60.02617615899442,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 64,
+    "upload_pages_per_commit": 66,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -11,3 +11,4 @@
 {"step": 761100, "loss": 0.5393986701965332, "lr": 3e-05, "grad_norm": 0.2306723284861559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111402662411, "tokens_per_second": 153846.61418262724, "data_wait_s": 8.323999963977258e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 761200, "loss": 0.5209561586380005, "lr": 3e-05, "grad_norm": 0.21673488926816759, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111417437559, "tokens_per_second": 154829.1961360877, "data_wait_s": 1.8564999891168554e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761300, "loss": 0.567022442817688, "lr": 3e-05, "grad_norm": 0.22644278490856187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111432132086, "tokens_per_second": 154186.37125312866, "data_wait_s": 9.705999900688767e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 761400, "loss": 0.5207698345184326, "lr": 3e-05, "grad_norm": 0.22618513057158895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111446784082, "tokens_per_second": 153802.0080439591, "data_wait_s": 9.147000127995852e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f6aab10a899d164418d7fbfa4f624cdbcab6c603361b7d23ca279b5d506e152
-size 27699
+oid sha256:944035c8ca3cb7f1e69b39d92a25d0acb98161de67e8ec1148f133c4e7220be3
+size 29817