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": 76366.01009603475,
-      "required_pages_per_second_upper_estimate": 0.006619212207241204
+      "required_token_positions_per_second": 76372.78661186634,
+      "required_pages_per_second_upper_estimate": 0.006619799578458564
     },
     {
       "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": 9545.751262004344,
-      "required_pages_per_second_upper_estimate": 0.003807942900113429
+      "required_token_positions_per_second": 9546.598326483292,
+      "required_pages_per_second_upper_estimate": 0.003808280806798824
     },
     {
       "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": 19091.502524008687,
-      "required_pages_per_second_upper_estimate": 0.004416483109960777
+      "required_token_positions_per_second": 19093.196652966584,
+      "required_pages_per_second_upper_estimate": 0.004416875016879535
     },
     {
       "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": 28637.253786013032,
-      "required_pages_per_second_upper_estimate": 0.0017438585075093603
+      "required_token_positions_per_second": 28639.794979449878,
+      "required_pages_per_second_upper_estimate": 0.0017440132528570526
     },
     {
       "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": 28637.253786013032,
-      "required_pages_per_second_upper_estimate": 0.00220457506943946
+      "required_token_positions_per_second": 28639.794979449878,
+      "required_pages_per_second_upper_estimate": 0.002204770697544702
     },
     {
       "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": 28637.253786013032,
-      "required_pages_per_second_upper_estimate": 0.3555788492992417
+      "required_token_positions_per_second": 28639.794979449878,
+      "required_pages_per_second_upper_estimate": 0.35561040241690006
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.04177303057811521,
-    "producer_p95_seconds": 0.1167212660000132,
-    "gpu_update_seconds": 0.965455703490109,
-    "producer_batches_per_second": 23.938890383593513,
-    "gpu_batches_per_second": 1.0357803018667908,
-    "producer_headroom": 23.11193825606488,
+    "producer_batch_seconds": 0.049767076289066736,
+    "producer_p95_seconds": 0.11792859699994551,
+    "gpu_update_seconds": 0.9653700391304642,
+    "producer_batches_per_second": 20.093605543383884,
+    "gpu_batches_per_second": 1.0358722142451489,
+    "producer_headroom": 19.39776476968861,
     "sustainable": true,
-    "buffer_seconds": 60.82370931987687,
+    "buffer_seconds": 60.81831246521924,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 74,
+    "upload_pages_per_commit": 73,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -10,3 +10,4 @@
 {"step": 41000, "loss": 0.6337492130696774, "lr": 0.0003, "grad_norm": 0.13548076983130877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5999586229, "tokens_per_second": 147891.59145646973, "data_wait_s": 8.796000201982679e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 41100, "loss": 0.6908429637551308, "lr": 0.0003, "grad_norm": 0.14318674346975743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6014175298, "tokens_per_second": 150714.2879058922, "data_wait_s": 9.341999884782126e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 41200, "loss": 0.6780079640448093, "lr": 0.0003, "grad_norm": 0.1282292300012621, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6028842649, "tokens_per_second": 151570.47499912896, "data_wait_s": 1.101900011235557e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 41300, "loss": 0.6396035701036453, "lr": 0.0003, "grad_norm": 0.13159565735574813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6043338175, "tokens_per_second": 149795.19507637958, "data_wait_s": 1.0010000096372096e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c7d0309d0452e226f25eb6ad1f6727e6be169dbfe2e59cd3e048bb14aaefbce
-size 25581
+oid sha256:015a4cb37581f694d555c2ed903db77326159dcf0f4c8c8147183b4ef6bd5656
+size 27699