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": 77618.16662684297,
-      "required_pages_per_second_upper_estimate": 0.005369804866204365
+      "required_token_positions_per_second": 77503.63984273428,
+      "required_pages_per_second_upper_estimate": 0.0053618816375409755
     },
     {
       "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": 9702.270828355371,
-      "required_pages_per_second_upper_estimate": 0.00406213818937835
+      "required_token_positions_per_second": 9687.954980341785,
+      "required_pages_per_second_upper_estimate": 0.004056144442763963
     },
     {
       "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": 19404.541656710742,
-      "required_pages_per_second_upper_estimate": 0.004041454294916862
+      "required_token_positions_per_second": 19375.90996068357,
+      "required_pages_per_second_upper_estimate": 0.0040354910677030025
     },
     {
       "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": 29106.812485066115,
-      "required_pages_per_second_upper_estimate": 0.0017449089649504126
+      "required_token_positions_per_second": 29063.864941025353,
+      "required_pages_per_second_upper_estimate": 0.0017423343252622717
     },
     {
       "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": 29106.812485066115,
-      "required_pages_per_second_upper_estimate": 0.0022629266433761147
+      "required_token_positions_per_second": 29063.864941025353,
+      "required_pages_per_second_upper_estimate": 0.0022595876607332276
     },
     {
       "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": 29106.812485066115,
-      "required_pages_per_second_upper_estimate": 0.3733509380981018
+      "required_token_positions_per_second": 29063.864941025353,
+      "required_pages_per_second_upper_estimate": 0.37280005311662695
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.046103101781259426,
-    "producer_p95_seconds": 0.11715749900008632,
-    "gpu_update_seconds": 0.9498807199924557,
-    "producer_batches_per_second": 21.690514550291986,
-    "gpu_batches_per_second": 1.0527637617573102,
-    "producer_headroom": 20.603401578038188,
+    "producer_batch_seconds": 0.05161314232814007,
+    "producer_p95_seconds": 0.12712263000003077,
+    "gpu_update_seconds": 0.9512843545103743,
+    "producer_batches_per_second": 19.37491024364135,
+    "gpu_batches_per_second": 1.0512103928322247,
+    "producer_headroom": 18.4310489848188,
     "sustainable": true,
-    "buffer_seconds": 60.792366079517166,
+    "buffer_seconds": 60.882198688663955,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 54,
+    "upload_pages_per_commit": 56,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -5,3 +5,4 @@
 {"step": 525500, "loss": 0.5804863572120667, "lr": 0.0003, "grad_norm": 0.17124735976777736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76915582593, "tokens_per_second": 152172.4227502477, "data_wait_s": 1.25609999486187e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 525600, "loss": 0.5729817748069763, "lr": 0.0003, "grad_norm": 0.17505624514239204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76930290681, "tokens_per_second": 154543.33481110237, "data_wait_s": 1.9686999621626455e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 525700, "loss": 0.5619977712631226, "lr": 0.0003, "grad_norm": 0.15766055232488896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76944945407, "tokens_per_second": 153994.17875249273, "data_wait_s": 1.0803999884956283e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 525800, "loss": 0.596474289894104, "lr": 0.0003, "grad_norm": 0.17673111523208587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76959784739, "tokens_per_second": 155924.93354150702, "data_wait_s": 1.1544000244612107e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c3c5a48c959975bbc646cb393d718d37234fc76eb4e0e0a5197522eb0f93f51
-size 14991
+oid sha256:061875f842d966cc5f6a1134569bb195467dd9eff5e4160c81ba418a7613bc0c
+size 17109