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": 77268.8306877342,
-      "required_pages_per_second_upper_estimate": 0.005345636995365766
+      "required_token_positions_per_second": 76599.1062338081,
+      "required_pages_per_second_upper_estimate": 0.00529930390366833
     },
     {
       "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": 9658.603835966775,
-      "required_pages_per_second_upper_estimate": 0.004043855731535738
+      "required_token_positions_per_second": 9574.888279226012,
+      "required_pages_per_second_upper_estimate": 0.004008805776107997
     },
     {
       "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": 19317.20767193355,
-      "required_pages_per_second_upper_estimate": 0.0040232649290398315
+      "required_token_positions_per_second": 19149.776558452024,
+      "required_pages_per_second_upper_estimate": 0.003988393443557019
     },
     {
       "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": 28975.811507900322,
-      "required_pages_per_second_upper_estimate": 0.0017370556563962335
+      "required_token_positions_per_second": 28724.664837678036,
+      "required_pages_per_second_upper_estimate": 0.001721999796995169
     },
     {
       "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": 28975.811507900322,
-      "required_pages_per_second_upper_estimate": 0.002252741893613877
+      "required_token_positions_per_second": 28724.664837678036,
+      "required_pages_per_second_upper_estimate": 0.0022332163446815508
     },
     {
       "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": 28975.811507900322,
-      "required_pages_per_second_upper_estimate": 0.3716705982209095
+      "required_token_positions_per_second": 28724.664837678036,
+      "required_pages_per_second_upper_estimate": 0.36844915839558284
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4831020580078871,
-    "producer_p95_seconds": 0.3245534649977344,
-    "gpu_update_seconds": 0.9541751744368474,
-    "producer_batches_per_second": 2.0699559925776057,
-    "gpu_batches_per_second": 1.0480255898401447,
-    "producer_headroom": 1.9751006202943346,
+    "producer_batch_seconds": 0.11320153203143946,
+    "producer_p95_seconds": 0.31704656800138764,
+    "gpu_update_seconds": 0.9625177580395724,
+    "producer_batches_per_second": 8.83380270615304,
+    "gpu_batches_per_second": 1.0389418705757392,
+    "producer_headroom": 8.502691975690333,
     "sustainable": true,
-    "buffer_seconds": 60.113035989521386,
+    "buffer_seconds": 60.63861875649306,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399501.3194533084,
-    "upload_bytes_per_second": 5663953.178054042,
+    "generation_bytes_per_second": 415048.6298797957,
+    "upload_bytes_per_second": 5735303.402426599,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.72368323993149,
+    "estimated_full_shards_per_hour": 89.05977413459209,
     "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": [
-    {
-      "seconds": 0.9807241070011514,
-      "bytes": 2284775,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.0974279740003112,
       "bytes": 5896433,
@@ -562,6 +556,12 @@
       "bytes": 9859192,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1245696650003083,
+      "bytes": 8444588,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -370,3 +370,4 @@
 {"step": 722000, "loss": 0.5707347393035889, "lr": 8e-05, "grad_norm": 0.2197074037151571, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105681078705, "tokens_per_second": 153069.99553446317, "data_wait_s": 1.587899896549061e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 722100, "loss": 0.5596104860305786, "lr": 8e-05, "grad_norm": 0.2129464747381036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105695631662, "tokens_per_second": 151893.08017769098, "data_wait_s": 1.365999924018979e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 722200, "loss": 0.5516434907913208, "lr": 8e-05, "grad_norm": 0.22077290950308204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105710086105, "tokens_per_second": 150220.9483061592, "data_wait_s": 1.7589998606126755e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 722300, "loss": 0.5085393190383911, "lr": 8e-05, "grad_norm": 0.22677287371391655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105724561000, "tokens_per_second": 150868.86048445228, "data_wait_s": 1.598599919816479e-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 sha256:7d4fef8c74c153180cfb875b61ce548725f08eeb721e9102a827644cc1530c1a
-size 793542
+oid sha256:4f4cc79fffba1291126c5cd9f53c3bd63d0fd1465d3432dd8bcb598dbd71b039
+size 795660