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": 75788.37646261582,
-      "required_pages_per_second_upper_estimate": 0.006569144387895512
+      "required_token_positions_per_second": 75857.7834574447,
+      "required_pages_per_second_upper_estimate": 0.006575160410296858
     },
     {
       "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": 9473.547057826978,
-      "required_pages_per_second_upper_estimate": 0.003779139563518022
+      "required_token_positions_per_second": 9482.222932180588,
+      "required_pages_per_second_upper_estimate": 0.0037826004995135582
     },
     {
       "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": 18947.094115653956,
-      "required_pages_per_second_upper_estimate": 0.004383076766189094
+      "required_token_positions_per_second": 18964.445864361176,
+      "required_pages_per_second_upper_estimate": 0.004387090787872162
     },
     {
       "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": 28420.641173480934,
-      "required_pages_per_second_upper_estimate": 0.0017306679358847017
+      "required_token_positions_per_second": 28446.668796541766,
+      "required_pages_per_second_upper_estimate": 0.0017322528815727218
     },
     {
       "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": 28420.641173480934,
-      "required_pages_per_second_upper_estimate": 0.0021878996309046502
+      "required_token_positions_per_second": 28446.668796541766,
+      "required_pages_per_second_upper_estimate": 0.002189903309376945
     },
     {
       "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": 28420.641173480934,
-      "required_pages_per_second_upper_estimate": 0.35288924560737217
+      "required_token_positions_per_second": 28446.668796541766,
+      "required_pages_per_second_upper_estimate": 0.3532124215769369
     }
   ],
   "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": 0.056308949289064714,
-    "producer_p95_seconds": 0.15129278400036128,
-    "gpu_update_seconds": 0.9728140836526279,
-    "producer_batches_per_second": 17.759166395850357,
-    "gpu_batches_per_second": 1.027945644295462,
-    "producer_headroom": 17.276367183813708,
+    "producer_batch_seconds": 0.055231121195301824,
+    "producer_p95_seconds": 0.14565746099970056,
+    "gpu_update_seconds": 0.9719239956617044,
+    "producer_batches_per_second": 18.105734201265207,
+    "gpu_batches_per_second": 1.0288870369119563,
+    "producer_headroom": 17.59739752928246,
     "sustainable": true,
-    "buffer_seconds": 60.314473186462926,
+    "buffer_seconds": 60.25928773102567,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -20,3 +20,4 @@
 {"step": 157000, "loss": 0.6236628387123346, "lr": 0.0003, "grad_norm": 0.15793851099557044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22977662403, "tokens_per_second": 150220.5779283792, "data_wait_s": 1.3168999885238009e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157100, "loss": 0.603167487308383, "lr": 0.0003, "grad_norm": 0.13626083682581658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22992238879, "tokens_per_second": 150001.94093199095, "data_wait_s": 1.3675999980478082e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157200, "loss": 0.6457713954150677, "lr": 0.0003, "grad_norm": 0.15602882021255543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23006794010, "tokens_per_second": 149698.9111962909, "data_wait_s": 1.2147000234108418e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 157300, "loss": 0.5872071832418442, "lr": 0.0003, "grad_norm": 0.15409774392771908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23021545116, "tokens_per_second": 151733.42046990996, "data_wait_s": 1.871799986474798e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb7cfba871c88c6436fc1dd7fcd1e95da10faeacd883dfe0574974e6595e2760
-size 46761
+oid sha256:564f29144f1d8d8e99ff185c6208812e9faff103f0e5f95387644c12b0e17939
+size 48879