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": 77189.6780943636,
-      "required_pages_per_second_upper_estimate": 0.005340161035297069
+      "required_token_positions_per_second": 77304.10549431221,
+      "required_pages_per_second_upper_estimate": 0.005348077388333662
     },
     {
       "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": 9648.70976179545,
-      "required_pages_per_second_upper_estimate": 0.0040397132892919674
+      "required_token_positions_per_second": 9663.013186789027,
+      "required_pages_per_second_upper_estimate": 0.004045701834647299
     },
     {
       "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": 19297.4195235909,
-      "required_pages_per_second_upper_estimate": 0.004019143579588622
+      "required_token_positions_per_second": 19326.026373578054,
+      "required_pages_per_second_upper_estimate": 0.00402510163202758
     },
     {
       "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": 28946.129285386345,
-      "required_pages_per_second_upper_estimate": 0.0017352762524786502
+      "required_token_positions_per_second": 28989.03956036708,
+      "required_pages_per_second_upper_estimate": 0.0017378486579435492
     },
     {
       "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": 28946.129285386345,
-      "required_pages_per_second_upper_estimate": 0.0022504342313717143
+      "required_token_positions_per_second": 28989.03956036708,
+      "required_pages_per_second_upper_estimate": 0.0022537703165091136
     },
     {
       "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": 28946.129285386345,
-      "required_pages_per_second_upper_estimate": 0.37128986654078766
+      "required_token_positions_per_second": 28989.03956036708,
+      "required_pages_per_second_upper_estimate": 0.3718402734747769
     }
   ],
   "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": 0.07610347653915994,
-    "producer_p95_seconds": 0.2057720299999346,
-    "gpu_update_seconds": 0.955153614060526,
-    "producer_batches_per_second": 13.140004182140592,
-    "gpu_batches_per_second": 1.0469520140837076,
-    "producer_headroom": 12.550722483342012,
+    "producer_batch_seconds": 0.08619689596875446,
+    "producer_p95_seconds": 0.199028091999935,
+    "gpu_update_seconds": 0.9537397726621992,
+    "producer_batches_per_second": 11.601345834570312,
+    "gpu_batches_per_second": 1.0485040350248511,
+    "producer_headroom": 11.064664938838641,
     "sustainable": true,
-    "buffer_seconds": 60.17467768581314,
+    "buffer_seconds": 60.085605677718554,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 227713.94292937298,
+    "generation_bytes_per_second": 245161.38841077618,
     "upload_bytes_per_second": 4914764.016490251,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.86211124335186,
+    "estimated_full_shards_per_hour": 52.60592688791718,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -105,3 +105,4 @@
 {"step": 853000, "loss": 0.5508236885070801, "lr": 3e-05, "grad_norm": 0.2984796562300817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124855451709, "tokens_per_second": 154012.14148950728, "data_wait_s": 1.1611000445554964e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 853100, "loss": 0.5127390027046204, "lr": 3e-05, "grad_norm": 0.29956521229592914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124870195117, "tokens_per_second": 153405.68505391295, "data_wait_s": 1.4670999007648788e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 853200, "loss": 0.6221808195114136, "lr": 3e-05, "grad_norm": 0.3335337318323358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124884759904, "tokens_per_second": 151276.74453297333, "data_wait_s": 1.5126001017051749e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 853300, "loss": 0.5039713978767395, "lr": 3e-05, "grad_norm": 0.29438514400012666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124899312019, "tokens_per_second": 151829.12695658917, "data_wait_s": 1.2165000953245908e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9546db87fa8c5ea404a8502030f4d26f002d8a119c51b0e6a22e073db7d981e
-size 228357
+oid sha256:696ef6f467a3fb0710dcbbaa3cfd20a4335809d8d7d0a906223d7200a9be1a67
+size 230475