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": 75727.503948841,
-      "required_pages_per_second_upper_estimate": 0.006563868112681485
+      "required_token_positions_per_second": 75960.98409585326,
+      "required_pages_per_second_upper_estimate": 0.006584105580074486
     },
     {
       "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": 9465.937993605125,
-      "required_pages_per_second_upper_estimate": 0.0037761041940342767
+      "required_token_positions_per_second": 9495.123011981657,
+      "required_pages_per_second_upper_estimate": 0.0037877465342195855
     },
     {
       "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": 18931.87598721025,
-      "required_pages_per_second_upper_estimate": 0.0043795563200035555
+      "required_token_positions_per_second": 18990.246023963315,
+      "required_pages_per_second_upper_estimate": 0.004393059200728816
     },
     {
       "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": 28397.813980815372,
-      "required_pages_per_second_upper_estimate": 0.001729277879616393
+      "required_token_positions_per_second": 28485.36903594497,
+      "required_pages_per_second_upper_estimate": 0.001734609523107913
     },
     {
       "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": 28397.813980815372,
-      "required_pages_per_second_upper_estimate": 0.0021861423304235417
+      "required_token_positions_per_second": 28485.36903594497,
+      "required_pages_per_second_upper_estimate": 0.0021928825609353227
     },
     {
       "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": 28397.813980815372,
-      "required_pages_per_second_upper_estimate": 0.35260580827216526
+      "required_token_positions_per_second": 28485.36903594497,
+      "required_pages_per_second_upper_estimate": 0.35369294902895526
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.0559557438515661,
-    "producer_p95_seconds": 0.1513643759999468,
-    "gpu_update_seconds": 0.9735960668901514,
-    "producer_batches_per_second": 17.871266310974292,
-    "gpu_batches_per_second": 1.0271200079866671,
-    "producer_headroom": 17.399394590711037,
+    "producer_batch_seconds": 0.049721900914043715,
+    "producer_p95_seconds": 0.13672532099963064,
+    "gpu_update_seconds": 0.9706035391400998,
+    "producer_batches_per_second": 20.111861807712078,
+    "gpu_batches_per_second": 1.0302867851542596,
+    "producer_headroom": 19.520644249261945,
     "sustainable": true,
-    "buffer_seconds": 60.36295614718939,
+    "buffer_seconds": 60.177419426686185,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "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

@@ -30,3 +30,4 @@
 {"step": 278000, "loss": 0.5908274631947279, "lr": 0.0003, "grad_norm": 0.14033135135848648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40691426002, "tokens_per_second": 151521.56213670966, "data_wait_s": 1.3419999959296547e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278100, "loss": 0.5920380316674709, "lr": 0.0003, "grad_norm": 0.1494044021047503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40706113567, "tokens_per_second": 150710.19329023614, "data_wait_s": 1.1837999863928417e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278200, "loss": 0.6077223289757967, "lr": 0.0003, "grad_norm": 0.1378774455084411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40720756769, "tokens_per_second": 151397.99093341763, "data_wait_s": 1.0745000054157572e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 278300, "loss": 0.6166594307869673, "lr": 0.0003, "grad_norm": 0.17162817212766507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40735477528, "tokens_per_second": 151484.95224747804, "data_wait_s": 1.0196999937761575e-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 sha256:7ad3cd8c9c632c609c712e79cd2521b9f297c68ee5027dd8f3c938b3322c2da2
-size 67941
+oid sha256:fef5f0327196f7c79511aed8a5e6dad502aeb73dc21d2a0954c88626bacd6fde
+size 70059