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": 75833.71175771797,
-      "required_pages_per_second_upper_estimate": 0.006573073936373711
+      "required_token_positions_per_second": 76015.49851257422,
+      "required_pages_per_second_upper_estimate": 0.006588830751550333
     },
     {
       "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": 9479.213969714747,
-      "required_pages_per_second_upper_estimate": 0.0037814001793979365
+      "required_token_positions_per_second": 9501.937314071778,
+      "required_pages_per_second_upper_estimate": 0.0037904648612062304
     },
     {
       "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": 18958.427939429494,
-      "required_pages_per_second_upper_estimate": 0.004385698646851079
+      "required_token_positions_per_second": 19003.874628143556,
+      "required_pages_per_second_upper_estimate": 0.004396211938450674
     },
     {
       "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": 28437.64190914424,
-      "required_pages_per_second_upper_estimate": 0.0017317031914906304
+      "required_token_positions_per_second": 28505.811942215336,
+      "required_pages_per_second_upper_estimate": 0.001735854388844138
     },
     {
       "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": 28437.64190914424,
-      "required_pages_per_second_upper_estimate": 0.00218920839459706
+      "required_token_positions_per_second": 28505.811942215336,
+      "required_pages_per_second_upper_estimate": 0.0021944563124496022
     },
     {
       "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": 28437.64190914424,
-      "required_pages_per_second_upper_estimate": 0.3531003378465083
+      "required_token_positions_per_second": 28505.811942215336,
+      "required_pages_per_second_upper_estimate": 0.3539467815068271
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.050841793640614696,
-    "producer_p95_seconds": 0.14711638700009644,
-    "gpu_update_seconds": 0.9722325109913447,
-    "producer_batches_per_second": 19.668857614833545,
-    "gpu_batches_per_second": 1.0285605435888396,
-    "producer_headroom": 19.122702827200847,
+    "producer_batch_seconds": 0.04843985278122176,
+    "producer_p95_seconds": 0.13934757800006992,
+    "gpu_update_seconds": 0.9699074720637945,
+    "producer_batches_per_second": 20.644158530301333,
+    "gpu_batches_per_second": 1.03102618425258,
+    "producer_headroom": 20.022923613008786,
     "sustainable": true,
-    "buffer_seconds": 60.27841568146337,
+    "buffer_seconds": 60.134263267955255,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -20,3 +20,4 @@
 {"step": 357000, "loss": 0.6048266664147377, "lr": 0.0003, "grad_norm": 0.15053552171825507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52254457624, "tokens_per_second": 149857.96183073396, "data_wait_s": 1.0407999980088789e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357100, "loss": 0.5407655872404575, "lr": 0.0003, "grad_norm": 0.15601018103178194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52269172951, "tokens_per_second": 150804.18442888864, "data_wait_s": 1.0480000128154643e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357200, "loss": 0.6043527461588383, "lr": 0.0003, "grad_norm": 0.16037414751564252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52283949229, "tokens_per_second": 151956.113868134, "data_wait_s": 1.650200010772096e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 357300, "loss": 0.5804549641907215, "lr": 0.0003, "grad_norm": 0.16365984136978362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52298628386, "tokens_per_second": 151215.89184892506, "data_wait_s": 1.5361000350821996e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ec3707d93d11c8c0594f5c6256282fbbaec6ca83ddf68364d1001757cdbfc66
-size 46761
+oid sha256:f553450eaaa6458376d420d404c20ad2825e2ae8e492aad2a45e9ed7a19d84ae
+size 48879