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": 75503.65084529063,
-      "required_pages_per_second_upper_estimate": 0.006544465093016243
+      "required_token_positions_per_second": 75961.07047044772,
+      "required_pages_per_second_upper_estimate": 0.006584113066805419
     },
     {
       "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": 9437.956355661328,
-      "required_pages_per_second_upper_estimate": 0.003764941900295727
+      "required_token_positions_per_second": 9495.133808805966,
+      "required_pages_per_second_upper_estimate": 0.0037877508412342293
     },
     {
       "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": 18875.912711322657,
-      "required_pages_per_second_upper_estimate": 0.004366610201046978
+      "required_token_positions_per_second": 18990.26761761193,
+      "required_pages_per_second_upper_estimate": 0.004393064196039388
     },
     {
       "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": 28313.869066983985,
-      "required_pages_per_second_upper_estimate": 0.0017241660747889867
+      "required_token_positions_per_second": 28485.401426417895,
+      "required_pages_per_second_upper_estimate": 0.0017346114955177744
     },
     {
       "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": 28313.869066983985,
-      "required_pages_per_second_upper_estimate": 0.0021796800185824076
+      "required_token_positions_per_second": 28485.401426417895,
+      "required_pages_per_second_upper_estimate": 0.0021928850544435914
     },
     {
       "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": 28313.869066983985,
-      "required_pages_per_second_upper_estimate": 0.3515634933879333
+      "required_token_positions_per_second": 28485.401426417895,
+      "required_pages_per_second_upper_estimate": 0.35369335121022505
     }
   ],
   "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": 0.04061997438284948,
-    "producer_p95_seconds": 0.13761248599985265,
-    "gpu_update_seconds": 0.9764825829557701,
-    "producer_batches_per_second": 24.618430099803778,
-    "gpu_batches_per_second": 1.024083805952835,
-    "producer_headroom": 24.03946821217247,
+    "producer_batch_seconds": 0.041857517179693104,
+    "producer_p95_seconds": 0.13717540899961023,
+    "gpu_update_seconds": 0.9706024354762551,
+    "producer_batches_per_second": 23.890571332910863,
+    "gpu_batches_per_second": 1.030287956684675,
+    "producer_headroom": 23.188246720642486,
     "sustainable": true,
-    "buffer_seconds": 60.54192014325774,
+    "buffer_seconds": 60.17735099952782,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 57,
+    "upload_pages_per_commit": 54,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -4,3 +4,4 @@
 {"step": 85400, "loss": 0.6122558545321226, "lr": 0.0003, "grad_norm": 0.1413275398458881, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12497887970, "tokens_per_second": 148746.17292177535, "data_wait_s": 6.886999472044408e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 85500, "loss": 0.6546185966581106, "lr": 0.0003, "grad_norm": 0.14180165680661713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12512511304, "tokens_per_second": 149284.1533241769, "data_wait_s": 1.4850000297883525e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 85600, "loss": 0.6194189470261335, "lr": 0.0003, "grad_norm": 0.13542036466516538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12527092704, "tokens_per_second": 149531.09426288542, "data_wait_s": 7.360000381595455e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 85700, "loss": 0.6051968038082123, "lr": 0.0003, "grad_norm": 0.13198794077910492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12541770395, "tokens_per_second": 150425.11568733782, "data_wait_s": 1.1227000868530013e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd7f546b3a9a3933b573fa34d2723c96d03068b01d3e53bbd920b4dc15aa5f02
-size 12873
+oid sha256:2f9c0d174fd31fc9c751c9f345aee8c90e712905e6544267548ec28a5d07f558
+size 14991