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": 76504.70198814316,
-      "required_pages_per_second_upper_estimate": 0.006631233668937775
+      "required_token_positions_per_second": 76338.24530789551,
+      "required_pages_per_second_upper_estimate": 0.006616805625774508
     },
     {
       "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": 9563.087748517895,
-      "required_pages_per_second_upper_estimate": 0.003814858683787257
+      "required_token_positions_per_second": 9542.28066348694,
+      "required_pages_per_second_upper_estimate": 0.003806558426474764
     },
     {
       "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": 19126.17549703579,
-      "required_pages_per_second_upper_estimate": 0.004424504092047114
+      "required_token_positions_per_second": 19084.56132697388,
+      "required_pages_per_second_upper_estimate": 0.00441487738618767
     },
     {
       "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": 28689.263245553688,
-      "required_pages_per_second_upper_estimate": 0.0017470256107228407
+      "required_token_positions_per_second": 28626.841990460816,
+      "required_pages_per_second_upper_estimate": 0.0017432244837866998
     },
     {
       "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": 28689.263245553688,
-      "required_pages_per_second_upper_estimate": 0.0022085789016062984
+      "required_token_positions_per_second": 28626.841990460816,
+      "required_pages_per_second_upper_estimate": 0.002203773540596156
     },
     {
       "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": 28689.263245553688,
-      "required_pages_per_second_upper_estimate": 0.35622463272227284
+      "required_token_positions_per_second": 28626.841990460816,
+      "required_pages_per_second_upper_estimate": 0.35544956964452135
     }
   ],
   "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": 0.05332555141407269,
-    "producer_p95_seconds": 0.12975135800002136,
-    "gpu_update_seconds": 0.9637054727881497,
-    "producer_batches_per_second": 18.75273623023612,
-    "gpu_batches_per_second": 1.0376614310457786,
-    "producer_headroom": 18.072114534831165,
+    "producer_batch_seconds": 0.04759646289844177,
+    "producer_p95_seconds": 0.12602733200014882,
+    "gpu_update_seconds": 0.9658068469170651,
+    "producer_batches_per_second": 21.00996458778323,
+    "gpu_batches_per_second": 1.0354037178262738,
+    "producer_headroom": 20.291567652366115,
     "sustainable": true,
-    "buffer_seconds": 60.71344478565343,
+    "buffer_seconds": 60.845831355775104,
     "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

@@ -12,3 +12,4 @@
 {"step": 41200, "loss": 0.6780079640448093, "lr": 0.0003, "grad_norm": 0.1282292300012621, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6028842649, "tokens_per_second": 151570.47499912896, "data_wait_s": 1.101900011235557e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 41300, "loss": 0.6396035701036453, "lr": 0.0003, "grad_norm": 0.13159565735574813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6043338175, "tokens_per_second": 149795.19507637958, "data_wait_s": 1.0010000096372096e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 41400, "loss": 0.6245260834693909, "lr": 0.0003, "grad_norm": 0.12509892006372614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6057938636, "tokens_per_second": 150838.87106638926, "data_wait_s": 6.7720000060944585e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 41500, "loss": 0.6294730324298143, "lr": 0.0003, "grad_norm": 0.13219610104495547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6072666804, "tokens_per_second": 151918.62530588364, "data_wait_s": 9.390999821334844e-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 sha256:6284d38d90d6811f42ba7965e41b51f7ca04cdeac2b692bbb10016ea8e77a5ca
-size 29817
+oid sha256:62c75b20482e7a632d939f5af20fe1dbdb6690fc8bcc7594d4142ea7852f75b9
+size 31935