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": 76076.50108351244,
-      "required_pages_per_second_upper_estimate": 0.005263148866759863
+      "required_token_positions_per_second": 75878.7739407829,
+      "required_pages_per_second_upper_estimate": 0.005249469644235653
     },
     {
       "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": 9509.562635439055,
-      "required_pages_per_second_upper_estimate": 0.003981455293208964
+      "required_token_positions_per_second": 9484.846742597863,
+      "required_pages_per_second_upper_estimate": 0.00397110726500289
     },
     {
       "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": 19019.12527087811,
-      "required_pages_per_second_upper_estimate": 0.003961182225861526
+      "required_token_positions_per_second": 18969.693485195727,
+      "required_pages_per_second_upper_estimate": 0.003950886888507738
     },
     {
       "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": 28528.68790631717,
-      "required_pages_per_second_upper_estimate": 0.0017102512791995325
+      "required_token_positions_per_second": 28454.54022779359,
+      "required_pages_per_second_upper_estimate": 0.0017058062390889936
     },
     {
       "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": 28528.68790631717,
-      "required_pages_per_second_upper_estimate": 0.002217980000279664
+      "required_token_positions_per_second": 28454.54022779359,
+      "required_pages_per_second_upper_estimate": 0.002212215344415626
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28528.68790631717,
-      "required_pages_per_second_upper_estimate": 0.3659353767437202
+      "required_token_positions_per_second": 28454.54022779359,
+      "required_pages_per_second_upper_estimate": 0.3649842899371941
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09052958598425676,
+    "producer_batch_seconds": 0.10491167495305831,
     "producer_p95_seconds": 0.27030788699994446,
-    "gpu_update_seconds": 0.9691297437439401,
-    "producer_batches_per_second": 11.046112595432632,
-    "gpu_batches_per_second": 1.0318535845745502,
-    "producer_headroom": 10.705116268978335,
+    "gpu_update_seconds": 0.9716551305578368,
+    "producer_batches_per_second": 9.531827610677649,
+    "gpu_batches_per_second": 1.0291717385631363,
+    "producer_headroom": 9.261649201507785,
     "sustainable": true,
-    "buffer_seconds": 60.086044112124284,
+    "buffer_seconds": 60.24261809458588,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 161,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 510082.6641006888,
+    "generation_bytes_per_second": 674221.5798409348,
     "upload_bytes_per_second": 3982873.927649975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.45186559930323,
+    "estimated_full_shards_per_hour": 72.33612797937886,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 823400, "loss": 0.5219301581382751, "lr": 3e-05, "grad_norm": 0.28984945960498276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120522630604, "tokens_per_second": 150153.22537021586, "data_wait_s": 1.5391000488307327e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823500, "loss": 0.5565692186355591, "lr": 3e-05, "grad_norm": 0.27770056215623146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120537270535, "tokens_per_second": 151278.89193235245, "data_wait_s": 1.1182997695868835e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823600, "loss": 0.5208242535591125, "lr": 3e-05, "grad_norm": 0.2910052313811568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120551893798, "tokens_per_second": 151255.03495848106, "data_wait_s": 1.5976998838596046e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 823700, "loss": 0.5073413848876953, "lr": 3e-05, "grad_norm": 0.28283460045958064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120566547181, "tokens_per_second": 151235.4149206287, "data_wait_s": 1.0415998985990882e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:65cf2771f9719a79b586f3ef542fbaa57f30a4d56054bc1e920709774537f16c
-size 503145
+oid sha256:06e618c68dfd7b6b3f0a0d9f9353672f61d932671a1eb9bbe86ff27cbd14fdf9
+size 505263