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": 76167.73500336225,
-      "required_pages_per_second_upper_estimate": 0.00660202622473198
+      "required_token_positions_per_second": 76108.58541228583,
+      "required_pages_per_second_upper_estimate": 0.006596899288090744
     },
     {
       "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": 9520.966875420281,
-      "required_pages_per_second_upper_estimate": 0.003798056037745445
+      "required_token_positions_per_second": 9513.573176535729,
+      "required_pages_per_second_upper_estimate": 0.0037951065807147475
     },
     {
       "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": 19041.933750840562,
-      "required_pages_per_second_upper_estimate": 0.004405016246668943
+      "required_token_positions_per_second": 19027.146353071457,
+      "required_pages_per_second_upper_estimate": 0.004401595442444373
     },
     {
       "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": 28562.90062626084,
-      "required_pages_per_second_upper_estimate": 0.0017393307901813326
+      "required_token_positions_per_second": 28540.719529607188,
+      "required_pages_per_second_upper_estimate": 0.0017379800777703451
     },
     {
       "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": 28562.90062626084,
-      "required_pages_per_second_upper_estimate": 0.0021988511573790173
+      "required_token_positions_per_second": 28540.719529607188,
+      "required_pages_per_second_upper_estimate": 0.0021971435951574133
     },
     {
       "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": 28562.90062626084,
-      "required_pages_per_second_upper_estimate": 0.35465563189913757
+      "required_token_positions_per_second": 28540.719529607188,
+      "required_pages_per_second_upper_estimate": 0.35438021691405425
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.05835703689844074,
-    "producer_p95_seconds": 0.1394400740000492,
-    "gpu_update_seconds": 0.9679689175048392,
-    "producer_batches_per_second": 17.13589402663313,
-    "gpu_batches_per_second": 1.0330910238086242,
-    "producer_headroom": 16.58701279143771,
+    "producer_batch_seconds": 0.05422725875000012,
+    "producer_p95_seconds": 0.14343000600001687,
+    "gpu_update_seconds": 0.9687211974918464,
+    "producer_batches_per_second": 18.44091003401491,
+    "gpu_batches_per_second": 1.032288756134519,
+    "producer_headroom": 17.86410045099033,
     "sustainable": true,
-    "buffer_seconds": 60.014072885300024,
+    "buffer_seconds": 60.06071424449448,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 65,
+    "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -14,3 +14,4 @@
 {"step": 356400, "loss": 0.6035982519388199, "lr": 0.0003, "grad_norm": 0.14425823427770984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52166934931, "tokens_per_second": 151385.60718366815, "data_wait_s": 1.2531000038507045e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356500, "loss": 0.6198965087532997, "lr": 0.0003, "grad_norm": 0.17267438068680266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52181420610, "tokens_per_second": 149105.94934928452, "data_wait_s": 1.1697999980242457e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 356600, "loss": 0.6039838008582592, "lr": 0.0003, "grad_norm": 0.15690151570082372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52196046402, "tokens_per_second": 150427.10282492568, "data_wait_s": 1.3294999916979577e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 356700, "loss": 0.6328282952308655, "lr": 0.0003, "grad_norm": 0.19027880083121018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52210646056, "tokens_per_second": 150503.8607225417, "data_wait_s": 7.793999884597724e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8fa6d5ebf68820d91f64159b41989b2583ef115149fc9d7f21b88e20f60ab933
-size 34053
+oid sha256:ebcfefbaccac99076050d8b6e7790b6fadd09af062d98fb225dd1a145d770c81
+size 36171