Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76340.7533861619,
-      "required_pages_per_second_upper_estimate": 0.006617023019641972
+      "required_token_positions_per_second": 76274.23098271122,
+      "required_pages_per_second_upper_estimate": 0.006611257026310357
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9542.594173270238,
-      "required_pages_per_second_upper_estimate": 0.003806683490214711
+      "required_token_positions_per_second": 9534.278872838902,
+      "required_pages_per_second_upper_estimate": 0.003803366392547831
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19085.188346540475,
-      "required_pages_per_second_upper_estimate": 0.0044150224361292345
+      "required_token_positions_per_second": 19068.557745677805,
+      "required_pages_per_second_upper_estimate": 0.004411175239308234
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28627.782519810713,
-      "required_pages_per_second_upper_estimate": 0.0017432817570895323
+      "required_token_positions_per_second": 28602.836618516707,
+      "required_pages_per_second_upper_estimate": 0.0017417626825817565
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28627.782519810713,
-      "required_pages_per_second_upper_estimate": 0.0022038459451490626
+      "required_token_positions_per_second": 28602.836618516707,
+      "required_pages_per_second_upper_estimate": 0.002201925540612776
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28627.782519810713,
-      "required_pages_per_second_upper_estimate": 0.35546124787129785
+      "required_token_positions_per_second": 28602.836618516707,
+      "required_pages_per_second_upper_estimate": 0.35515150326578726
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.0453818802343946,
-    "producer_p95_seconds": 0.12252325299982658,
-    "gpu_update_seconds": 0.9657751165626365,
-    "producer_batches_per_second": 22.03522628051244,
-    "gpu_batches_per_second": 1.035437735814913,
-    "producer_headroom": 21.281073229545974,
+    "producer_batch_seconds": 0.04752139327344018,
+    "producer_p95_seconds": 0.12040669899988643,
+    "gpu_update_seconds": 0.9666174152147353,
+    "producer_batches_per_second": 21.04315406423284,
+    "gpu_batches_per_second": 1.0345354679729712,
+    "producer_headroom": 20.340679189534203,
     "sustainable": true,
-    "buffer_seconds": 60.843832343446095,
+    "buffer_seconds": 60.896897158528326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 40700, "loss": 0.6462512910366058, "lr": 0.0003, "grad_norm": 0.1260881089485152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5955780384, "tokens_per_second": 152464.35437724792, "data_wait_s": 8.34499996926752e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 40800, "loss": 0.6299891471862793, "lr": 0.0003, "grad_norm": 0.1282002006955341, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5970591672, "tokens_per_second": 153031.50164184143, "data_wait_s": 1.12830000489339e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 40900, "loss": 0.6571875438094139, "lr": 0.0003, "grad_norm": 0.14368758349841199, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5985111871, "tokens_per_second": 149894.62211921526, "data_wait_s": 1.1872999948536744e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 41000, "loss": 0.6337492130696774, "lr": 0.0003, "grad_norm": 0.13548076983130877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5999586229, "tokens_per_second": 147891.59145646973, "data_wait_s": 8.796000201982679e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3174d1a83d7e22af1642dde0a435a68ab80d620fa2580e33066bd50e34b3e9e
-size 19227
+oid sha256:ae4a6671c4a65a99a04066a8a3d27998d02f605c6e06b9373df84e7807b1f66f
+size 21345