Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77372.46190130063,
-      "required_pages_per_second_upper_estimate": 0.005352806443177836
+      "required_token_positions_per_second": 77335.35201954897,
+      "required_pages_per_second_upper_estimate": 0.005350239095451468
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9671.557737662579,
-      "required_pages_per_second_upper_estimate": 0.004049279259667543
+      "required_token_positions_per_second": 9666.919002443621,
+      "required_pages_per_second_upper_estimate": 0.004047337118099172
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19343.115475325158,
-      "required_pages_per_second_upper_estimate": 0.004028660841251582
+      "required_token_positions_per_second": 19333.838004887242,
+      "required_pages_per_second_upper_estimate": 0.004026728588822713
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29014.673212987735,
-      "required_pages_per_second_upper_estimate": 0.0017393853562778362
+      "required_token_positions_per_second": 29000.757007330863,
+      "required_pages_per_second_upper_estimate": 0.0017385511009975225
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29014.673212987735,
-      "required_pages_per_second_upper_estimate": 0.0022557632202498475
+      "required_token_positions_per_second": 29000.757007330863,
+      "required_pages_per_second_upper_estimate": 0.0022546812964709456
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29014.673212987735,
-      "required_pages_per_second_upper_estimate": 0.37216907444732283
+      "required_token_positions_per_second": 29000.757007330863,
+      "required_pages_per_second_upper_estimate": 0.371990572303214
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.05502280687498029,
-    "producer_p95_seconds": 0.12825018700004875,
-    "gpu_update_seconds": 0.9528971702367485,
-    "producer_batches_per_second": 18.17428184411136,
-    "gpu_batches_per_second": 1.0494311781317902,
-    "producer_headroom": 17.318221740338828,
+    "producer_batch_seconds": 0.05368262117968392,
+    "producer_p95_seconds": 0.14249031899998954,
+    "gpu_update_seconds": 0.9533544242659283,
+    "producer_batches_per_second": 18.628002471281114,
+    "gpu_batches_per_second": 1.0489278431470943,
+    "producer_headroom": 17.759088571232496,
     "sustainable": true,
-    "buffer_seconds": 60.032521724915156,
+    "buffer_seconds": 60.06132872875348,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 61,
+    "upload_pages_per_commit": 65,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -11,3 +11,4 @@
 {"step": 646100, "loss": 0.5347440242767334, "lr": 0.0001, "grad_norm": 0.17545354614213696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94572621539, "tokens_per_second": 154321.7722860928, "data_wait_s": 1.0644999974829261e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 646200, "loss": 0.5362868309020996, "lr": 0.0001, "grad_norm": 0.19484272820118434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94587097212, "tokens_per_second": 151959.3742421955, "data_wait_s": 1.7045000049620285e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 646300, "loss": 0.5540004372596741, "lr": 0.0001, "grad_norm": 0.1673512198936814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94601667259, "tokens_per_second": 153199.93679297174, "data_wait_s": 1.202599992211617e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 646400, "loss": 0.507813036441803, "lr": 0.0001, "grad_norm": 0.16432517602460556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94616241350, "tokens_per_second": 152966.47065197848, "data_wait_s": 1.4149999969959026e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:901d469474baa03b7917c580cffb277f8853549996142f644f8d0f7343e6dcf3
-size 27699
+oid sha256:e4c21d62c83b3541c7ad3416e997494e3f3560b701808ddf7d101c9954d5e7a6
+size 29817