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": 75562.37831414526,
-      "required_pages_per_second_upper_estimate": 0.006549555441173153
+      "required_token_positions_per_second": 75536.43039267068,
+      "required_pages_per_second_upper_estimate": 0.0065473063411041425
     },
     {
       "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": 9445.297289268157,
-      "required_pages_per_second_upper_estimate": 0.0037678703084682292
+      "required_token_positions_per_second": 9442.053799083835,
+      "required_pages_per_second_upper_estimate": 0.003766576431739203
     },
     {
       "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": 18890.594578536315,
-      "required_pages_per_second_upper_estimate": 0.004370006592634819
+      "required_token_positions_per_second": 18884.10759816767,
+      "required_pages_per_second_upper_estimate": 0.0043685059438934885
     },
     {
       "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": 28335.89186780447,
-      "required_pages_per_second_upper_estimate": 0.0017255071478142479
+      "required_token_positions_per_second": 28326.161397251504,
+      "required_pages_per_second_upper_estimate": 0.001724914613209406
     },
     {
       "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": 28335.89186780447,
-      "required_pages_per_second_upper_estimate": 0.0021813753947525828
+      "required_token_positions_per_second": 28326.161397251504,
+      "required_pages_per_second_upper_estimate": 0.002180626316193749
     },
     {
       "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": 28335.89186780447,
-      "required_pages_per_second_upper_estimate": 0.3518369428685507
+      "required_token_positions_per_second": 28326.161397251504,
+      "required_pages_per_second_upper_estimate": 0.35171612299007293
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05149149152340371,
-    "producer_p95_seconds": 0.13285207299986723,
-    "gpu_update_seconds": 0.975723655672682,
-    "producer_batches_per_second": 19.420684280343362,
-    "gpu_batches_per_second": 1.0248803482278817,
-    "producer_headroom": 18.949221061681616,
+    "producer_batch_seconds": 0.05591037432029822,
+    "producer_p95_seconds": 0.14182022499971936,
+    "gpu_update_seconds": 0.976058831701873,
+    "producer_batches_per_second": 17.88576828821106,
+    "gpu_batches_per_second": 1.0245284070186453,
+    "producer_headroom": 17.4575620994817,
     "sustainable": true,
-    "buffer_seconds": 60.49486665170629,
+    "buffer_seconds": 60.515647565516126,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -19,3 +19,4 @@
 {"step": 196900, "loss": 0.5936473049223423, "lr": 0.0003, "grad_norm": 0.14001345759618117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28818393057, "tokens_per_second": 149003.66544643656, "data_wait_s": 2.1523000214074273e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 197000, "loss": 0.6025869958102703, "lr": 0.0003, "grad_norm": 0.1431011205334915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28832906369, "tokens_per_second": 148115.18466128802, "data_wait_s": 2.374300038354704e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 197100, "loss": 0.6236733943223953, "lr": 0.0003, "grad_norm": 0.15140581211059811, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28847548768, "tokens_per_second": 149996.85108211773, "data_wait_s": 2.2203999833436683e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 197200, "loss": 0.5887098088860512, "lr": 0.0003, "grad_norm": 0.13839620821846402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28862254588, "tokens_per_second": 150090.3536678603, "data_wait_s": 1.8150999949284596e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a25522e1adc2d9c5371b8fb5bb0b559cb6cfa75088c6bff2e1f94f11cd5a813
-size 44643
+oid sha256:bea00a701a8e207dd9520b13f6c0929decf7af4d0428a3aa42e210b515378c7d
+size 46761