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": 76220.30072168965,
-      "required_pages_per_second_upper_estimate": 0.006606582487970004
+      "required_token_positions_per_second": 76343.72095047656,
+      "required_pages_per_second_upper_estimate": 0.006617280240595544
     },
     {
       "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": 9527.537590211206,
-      "required_pages_per_second_upper_estimate": 0.0038006771941164856
+      "required_token_positions_per_second": 9542.96511880957,
+      "required_pages_per_second_upper_estimate": 0.003806831465936481
     },
     {
       "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": 19055.075180422413,
-      "required_pages_per_second_upper_estimate": 0.004408056285121443
+      "required_token_positions_per_second": 19085.93023761914,
+      "required_pages_per_second_upper_estimate": 0.004415194059573447
     },
     {
       "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": 28582.61277063362,
-      "required_pages_per_second_upper_estimate": 0.001740531156352006
+      "required_token_positions_per_second": 28628.895356428708,
+      "required_pages_per_second_upper_estimate": 0.0017433495230009658
     },
     {
       "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": 28582.61277063362,
-      "required_pages_per_second_upper_estimate": 0.0022003686528195418
+      "required_token_positions_per_second": 28628.895356428708,
+      "required_pages_per_second_upper_estimate": 0.002203931614392445
     },
     {
       "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": 28582.61277063362,
-      "required_pages_per_second_upper_estimate": 0.3549003907599441
+      "required_token_positions_per_second": 28628.895356428708,
+      "required_pages_per_second_upper_estimate": 0.3554750655776687
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.048148320492188645,
-    "producer_p95_seconds": 0.12625102499987406,
-    "gpu_update_seconds": 0.9673013528142584,
-    "producer_batches_per_second": 20.769156427008397,
-    "gpu_batches_per_second": 1.0338039919934034,
-    "producer_headroom": 20.090033108656176,
+    "producer_batch_seconds": 0.05026503760153744,
+    "producer_p95_seconds": 0.13791262899985668,
+    "gpu_update_seconds": 0.9657375758227275,
+    "producer_batches_per_second": 19.894543955725865,
+    "gpu_batches_per_second": 1.0354779859819412,
+    "producer_headroom": 19.21290865190139,
     "sustainable": true,
-    "buffer_seconds": 60.93998522729828,
+    "buffer_seconds": 60.84146727683183,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -24,3 +24,4 @@
 {"step": 277400, "loss": 0.567566778510809, "lr": 0.0003, "grad_norm": 0.14623475431320984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40603211126, "tokens_per_second": 150490.46349487503, "data_wait_s": 1.1923999863938661e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277500, "loss": 0.6089343391358852, "lr": 0.0003, "grad_norm": 0.1495398169538208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40617975581, "tokens_per_second": 152355.8195530897, "data_wait_s": 1.25619999380433e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277600, "loss": 0.5990518778562546, "lr": 0.0003, "grad_norm": 0.16147448335486667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40632621246, "tokens_per_second": 150908.6515825122, "data_wait_s": 1.2711000181298004e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 277700, "loss": 0.5739009454846382, "lr": 0.0003, "grad_norm": 0.13901132147585363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40647240294, "tokens_per_second": 151002.51795758354, "data_wait_s": 1.247000000148546e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8868c14b90bf5fa6fca21c1e88c18e51066723aed0b77e0ed4f89c7a6c899d41
-size 55233
+oid sha256:c6c4d6cce73621c871e9c4f5ecd6dd69bca8cf1d1386dbae97e02c60fef62601
+size 57351