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": 77581.76578345409,
-      "required_pages_per_second_upper_estimate": 0.005367286571397129
+      "required_token_positions_per_second": 77480.86716235138,
+      "required_pages_per_second_upper_estimate": 0.005360306170672188
     },
     {
       "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": 9697.720722931761,
-      "required_pages_per_second_upper_estimate": 0.004060233155254491
+      "required_token_positions_per_second": 9685.108395293923,
+      "required_pages_per_second_upper_estimate": 0.004054952637047878
     },
     {
       "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": 19395.441445863522,
-      "required_pages_per_second_upper_estimate": 0.0040395589609860845
+      "required_token_positions_per_second": 19370.216790587845,
+      "required_pages_per_second_upper_estimate": 0.0040343053305110545
     },
     {
       "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": 29093.162168795287,
-      "required_pages_per_second_upper_estimate": 0.0017440906493328026
+      "required_token_positions_per_second": 29055.325185881768,
+      "required_pages_per_second_upper_estimate": 0.001741822379980865
     },
     {
       "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": 29093.162168795287,
-      "required_pages_per_second_upper_estimate": 0.002261865391327454
+      "required_token_positions_per_second": 29055.325185881768,
+      "required_pages_per_second_upper_estimate": 0.002258923732333227
     },
     {
       "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": 29093.162168795287,
-      "required_pages_per_second_upper_estimate": 0.3731758464975473
+      "required_token_positions_per_second": 29055.325185881768,
+      "required_pages_per_second_upper_estimate": 0.3726905143069197
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.05542736330469822,
-    "producer_p95_seconds": 0.14089390499975707,
-    "gpu_update_seconds": 0.9503263976459274,
-    "producer_batches_per_second": 18.041630349665876,
-    "gpu_batches_per_second": 1.0522700437208943,
-    "producer_headroom": 17.145437577857404,
+    "producer_batch_seconds": 0.04876409598432119,
+    "producer_p95_seconds": 0.1429294250001476,
+    "gpu_update_seconds": 0.9515639499169811,
+    "producer_batches_per_second": 20.506890978180415,
+    "gpu_batches_per_second": 1.0509015185865802,
+    "producer_headroom": 19.51361817971426,
     "sustainable": true,
-    "buffer_seconds": 60.820889449339354,
+    "buffer_seconds": 60.90009279468679,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -37,3 +37,4 @@
 {"step": 688700, "loss": 0.5927730202674866, "lr": 8e-05, "grad_norm": 0.2079924289921846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100807133319, "tokens_per_second": 152234.86601582528, "data_wait_s": 1.4876000022923108e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688800, "loss": 0.5484572052955627, "lr": 8e-05, "grad_norm": 0.1872351369526101, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100821681614, "tokens_per_second": 152381.7719375346, "data_wait_s": 1.6102000245155068e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688900, "loss": 0.5259572863578796, "lr": 8e-05, "grad_norm": 0.1889048037181038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100836258126, "tokens_per_second": 152824.50179066346, "data_wait_s": 1.6406000213464722e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 689000, "loss": 0.5641706585884094, "lr": 8e-05, "grad_norm": 0.1990296048883631, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100850908471, "tokens_per_second": 153449.5390437188, "data_wait_s": 1.5900999642326497e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a58d498f0a9e8621201ceb1702b9a441e4b2cbf8e7dce6d8fd6953e7c19589ce
-size 82767
+oid sha256:5fd1f4a2e618f3e93d4c19f912ee016a564003b939461e79ccaea87c73705ef1
+size 84885