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": 77348.55352139378,
-      "required_pages_per_second_upper_estimate": 0.005351152405980796
+      "required_token_positions_per_second": 76682.23865861837,
+      "required_pages_per_second_upper_estimate": 0.005305055197710506
     },
     {
       "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": 9668.569190174223,
-      "required_pages_per_second_upper_estimate": 0.004048028017242137
+      "required_token_positions_per_second": 9585.279832327296,
+      "required_pages_per_second_upper_estimate": 0.0040131565023912
     },
     {
       "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": 19337.138380348446,
-      "required_pages_per_second_upper_estimate": 0.004027415969994528
+      "required_token_positions_per_second": 19170.559664654593,
+      "required_pages_per_second_upper_estimate": 0.0039927220164916455
     },
     {
       "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": 29005.70757052267,
-      "required_pages_per_second_upper_estimate": 0.001738847879701799
+      "required_token_positions_per_second": 28755.83949698189,
+      "required_pages_per_second_upper_estimate": 0.0017238686702194864
     },
     {
       "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": 29005.70757052267,
-      "required_pages_per_second_upper_estimate": 0.0022550661809838824
+      "required_token_positions_per_second": 28755.83949698189,
+      "required_pages_per_second_upper_estimate": 0.0022356400373195907
     },
     {
       "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": 29005.70757052267,
-      "required_pages_per_second_upper_estimate": 0.3720540728123378
+      "required_token_positions_per_second": 28755.83949698189,
+      "required_pages_per_second_upper_estimate": 0.36884903345239145
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0831242215391228,
-    "producer_p95_seconds": 0.2285701950022485,
-    "gpu_update_seconds": 0.9531917100377012,
-    "producer_batches_per_second": 12.03018784999202,
-    "gpu_batches_per_second": 1.0491068999755018,
-    "producer_headroom": 11.467075328808669,
+    "producer_batch_seconds": 0.3593596827656711,
+    "producer_p95_seconds": 0.24045200600085082,
+    "gpu_update_seconds": 0.9614742773516257,
+    "producer_batches_per_second": 2.7827273006918625,
+    "gpu_batches_per_second": 1.0400694262507917,
+    "producer_headroom": 2.6755207204993487,
     "sustainable": true,
-    "buffer_seconds": 60.05107773237518,
+    "buffer_seconds": 60.57287947315242,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426643.7952058317,
+    "generation_bytes_per_second": 385192.35711023485,
     "upload_bytes_per_second": 5746159.7853829805,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.54782669192518,
+    "estimated_full_shards_per_hour": 82.65331301670345,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

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

@@ -199,3 +199,4 @@
 {"step": 704900, "loss": 0.4992697238922119, "lr": 8e-05, "grad_norm": 0.2101155164146234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103177784453, "tokens_per_second": 152551.52564595608, "data_wait_s": 1.8695998733164743e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 705000, "loss": 0.5618807673454285, "lr": 8e-05, "grad_norm": 0.20961666563614817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103192342154, "tokens_per_second": 151723.58814000429, "data_wait_s": 1.3707998732570559e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 705100, "loss": 0.575244128704071, "lr": 8e-05, "grad_norm": 0.22217285645204765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103206859900, "tokens_per_second": 142009.79687112584, "data_wait_s": 1.3758999557467178e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 705200, "loss": 0.522087037563324, "lr": 8e-05, "grad_norm": 0.20907069136539252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103221483343, "tokens_per_second": 152179.45252962134, "data_wait_s": 1.9398001313675195e-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 sha256:4badc8bbc312d280427eaacce680d55e87830ddb6777ab325678cb2233aa0ad7
-size 429015
+oid sha256:23e8e86bef8d7321f57990351ead8c78eaa5daa46d681580c3ba216ac43a4f49
+size 431133